oxford textbooks in linguistics - 213.230.96.51:8090

oxford textbooks in linguistics

Series editors

Keith Brown, Eve V. Clark, April McMahon, Jim Miller, and Lesley Milroy

The Grammar of WordsSecond edition

Oxford Textbooks in L inguistics

General editors: Keith Brown, University of Cambridge; Eve V. Clark, Stanford University;

April McMahon, University of Edinburgh; Jim Miller, University of Auckland;

Lesley Milroy, University of Michigan

This series provides lively and authoritative introductions to the approaches, methods, and

theories associated with the main subWelds of linguistics.

published

The Grammar of Words

An Introduction to Linguistic Morphology

by Geert Booij

A Practical Introduction to Phonetics

Second edition

by J. C. Catford

Meaning in Language

An Introduction to Semantics and Pragmatics

Second edition

by Alan Cruse

Principles and Parameters

An Introduction to Syntactic Theory

by Peter W. Culicover

A Semantic Approach to English Grammer

by R. M. W. Dixon

Semantic Analysis

A Practical Introduction

by CliV Goddard

Pragmatics

by Yan Huang

Diachronic Syntax

by Ian Roberts

Cognitive Grammar

An Introduction

by John R. Taylor

Linguistic Categorization

Third edition

by John R. Taylor

in preparation

Natural Language Syntax

by Peter W. Culicover

Translation

Theroy and Practice

by Kirsten Malmkjaer

The Grammar ofWords

An Introduction to LinguisticMorphology

Second edition

Geert Booij

1

3Great Clarendon Street, Oxford ox2 6dp

Oxford University Press is a department of the University of Oxford.It furthers the University’s objective of excellence in research, scholarship,

and education by publishing worldwide in

Oxford New York

Auckland Cape Town Dar es Salaam Hong Kong KarachiKuala Lumpur Madrid Melbourne Mexico City Nairobi

New Delhi Shanghai Taipei Toronto

With oYces in

Argentina Austria Brazil Chile Czech Republic France GreeceGuatemala Hungary Italy Japan South Korea Poland Portugal

Singapore Switzerland Thailand Turkey Ukraine Vietnam

Oxford is a registered trade mark of Oxford University Pressin the UK and in certain other countries

Published in the United Statesby Oxford University Press Inc. New York

� Geert Booij 2005; 2007

The moral rights of the author have been assertedDatabase right Oxford University Press (maker)

First edition published 2005Second edition published 2007

All rights reserved. No part of this publication may be reproduced,stored in a retrieval system, or transmitted, in any form or by any means,

without the prior permission in writing of Oxford University Press,or as expressly permitted by law, or under terms agreed with the appropriate

reprographics rights organization. Enquiries concerning reproductionoutside the scope of the above should be sent to the Rights Department,

Oxford University Press, at the address above

You must not circulate this book in any other binding or coverand you must impose this same condition on any acquirer

British Library Cataloguing in Publication Data

Data applied for

Library of Congress Cataloging in Publication Data

Data available

Typeset bySPi Publisher Services, Pondicherry, India

Printed in Great Britain byBiddles Ltd. www.biddles.co.uk

ISBN: 978–0–19–922624–5

1 3 5 7 9 10 8 6 4 2

www.biddles.co.uk

Contents

Typographic Conventions vii

Abbreviations and Symbols viii

List of Figures xi

List of Tables xii

Preface to the Second Edition xiii

Preface to the First Edition xiv

I What is Linguistic Morphology? 1

1 Morphology: basic notions 3

2 Morphological analysis 27

II Word-Formation 49

3 Derivation 51

4 Compounding 75

III Inflection 97

5 InXection 99

6 InXectional systems 125

IV Interfaces 151

7 The interface between morphology and phonology 153

8 Morphology and syntax: demarcation and interaction 185

9 Morphology and semantics 207

V Morphology and Mind 229

10 Morphology and psycholinguistics 231

11 Morphology and language change 255

VI Conclusions 279

12 The word as a linguistic unit 281

Answers to questions 295

Glossary of terms 307

References 325

Language index 339

Index of terms 341

vi contents

Typographic Conventions

Small capitals For lexemes, semantic components, and

morphosyntactic categories

Bold type For technical terms Wrst introduced

Italics For citation formswhen not set on diVerent

lines

Single quotation marks For quotations from other authors

Double quotation marks For glosses

Questions marks For formal or semantic oddness

Asterisks For ungrammaticality

// For underlying phonological representa-

tions

[ ] For phonetic representations

Abbreviations and Symbols

A Adjective, transitive subject

ABL Ablative

ABS Absolutive

ACC Accusative

AFF,aV AYx

ALL Allative

ANTIPASS Antipassive

AP Adjective Phrase

APPL Applicative

AUX Auxiliary

CAUS Causative

COM Comitative

COMP Completive aspect

COORD Coordination

CVB Converb

DAT Dative

DO Direct Object

DS DiVerent Subject

DIM Diminutive

ERG Ergative

EVID Evidential

F Foot

FEM Feminine

FUT Future

GEN Genitive

H High Tone

IMP Imperative

IMPF Imperfective

INCOMP Incompletive aspect

IND Indicative

INF InWnitive

INSTR Instrumental

INTR Intransitive

IO Indirect Object

L Low Tone

LCS Lexical Conceptual Structure

LOC Locative

M Mid Tone

MASC Masculine

N Noun, Number of tokens

n1 number of hapaxes

NEG Negation

NEUT Neuter

NF Non-future

NOM Nominative

NOM1 Nominalized Verb Stem 1

NONFUT Non-future

NP Noun Phrase

O Transitive object

OBJ Object

P Preposition, Productivity

P* Global productivity

PART Partitive

PAS Predicate Argument Structure

PASS Passive

PERF Perfect(ive)

PERS Person

PL Plural

PP Prepositional Phrase

PRES Present

PRET Preterite

PROG Progressive

PTCP Participle

Q Qualifying

R Relation, Relational

REL Relative case

RHR Right-hand Head Rule

S Sentence, intransitive subject

s strong

SC Subject Concord

SG Singular

abbreviations and symbols ix

SUBJ Subject

SUPERL Superlative

t trace

TMA Tense–Mood–Aspect

V Verb, Vowel, number of types

VP Verb Phrase

w weak

WFR Word Formation Rule

X, Y, x, y variables

� syllable

ø phonological word

< derives from

> results in, changes to

� syllable boundary

- morpheme boundary

´ primary stress, high tone, long vowel

` secondary stress, low tone

) is transformed into

! is realized as, is changed to

, linked to

>> ranked higher than

[ ] phonetic form

// phonological form, underlying form

x abbreviations and symbols

List of Figures

1.1. The morphological structure of tranquillizer 10

1.2. The morphological structure of tranquillizer

without aYx labels 11

2.1. Three-tiered representations of words 38

3.1. Percolation of head features 53

3.2. Inheritance tree for -bar-adjectives 62

3.3. Non-redundant speciWcation of doable 63

4.1. Morphological tree of White House travel oYce staV 76

6.1. Stems in Latin 143

7.1. The representation of dog 154

7.2. The representation of baker 154

7.3. The template for deverbal -er 155

7.4. The morphological and phonological structure of hoeden 157

7.5. OT-tableaux for hoed and hoeden 159

7.6. The syllable structure of stump 160

7.7. OT-tableau for rekenaar 174

7.8. The prosodic structure of sugar cookie 176

7.9. Reduplication in Diyari 181

8.1. Syntactic derivation of noun–verb compounds 199

8.2. A syntactic account of causative verbs 201

9.1. The radial semantic structure of diminutives 223

10.1. Some association patterns for English past tense forms 246

10.2. Schema for English past tense forms 247

List of Tables

2.1. Distribution of Number morphemes 44

3.1. Productivity measure for the English suYxes -ity and -ness 70

5.1. Indicative forms of the Latin verb laudare 100

5.2. Imperfective and perfective 3 sg forms of laudare 114

6.1. InXection of Icelandic class I nouns 128

6.2. Noun classes in Swahili 130

6.3. The Italian verb paradigm 142

6.4. Latin verbal stem-forms 143

6.5. Singular present forms of ‘‘to be’’ 144

6.6. Regular past tense and participle formation in Dutch 144

6.7. Stem alternating verbs in Dutch 145

7.1. The singular paradigms of hestur and the deWnite article 167

7.2. Allomorph selection in Biak 175

10.1. Frequency data for calculate 234

Preface to the Second Edition

I would like to thank the following colleagues and students for their remarks

on the second edition of this book: Laurie Bauer, Jean-Marc Gachelin,

Alexandra Galani, Istvan Kenesei, Elizabeth Koier, Francesca Masini,

Franz Rainer, Iggy Roca, and Jeroen Wiedenhof. These remarks helped me

in correcting typos andminor infelicities in the Wrst edition. At the suggestion

of Alexandra Galani, I have added a glossary of terms to this edition.

Leiden, July 2006 G. B.

Preface to the First Edition

Each textbook provides a specific perspective on the discipline that it aims

to introduce. Therefore, writing this book has not only been a challenge for

me because of the didactic demands that each textbook imposes on its

writer. It also forced me to rethink my own ideas on morphology in

confrontation with those of others, and to come up with a consistent

picture of what morphology is about. This perspective is summarized by

the title of this book, The Grammar of Words, which gives the linguistic

entity of the word a pivotal role in understanding morphology.

It is with much pleasure that I would like to thank a number of col-

leagues for their constructive comments on an earlier draft of this book.

Andrew Carstairs-McCarthy (University of Canterbury at Christchurch,

New Zealand), Ingo Plag (University of Siegen), Sergio Scalise (University

of Bologna), Caro Struijke (Vrije Universiteit Amsterdam), and Greg

Stump (University of Kentucky at Lexington) read the whole manuscript,

and gave extremely valuable advice. Maarten Mous and Marian Klamer

(both University of Leiden) provided useful feedback for a number of

chapters, and Mirjam Ernestus (Max Planck Institute for Psycholinguis-

tics, Nijmegen) had a critical look at Chapter 10. Jenny Audring and

Lourens de Vries (Vrije Universiteit Amsterdam), Maarten Kossman (Uni-

versity of Leiden), and Jaap van Marle (Open Universiteit Heerlen) also

commented on a number of points. None of them should be held respon-

sible for what I wrote in this book.

Another form of support for this project came from my colleagues at the

University of Wisconsin at Madison, in particular Mark Louden, Monica

Macaulay, and Joe Salmons. They made it possible for me to spend a very

pleasant month in Madison, in which I could work on parts of this book.

It is my sincere hope that this textbook will prove to be useful for a new

generation of students of language, and that they will enjoy reading and

thinking about the many wonderful intricacies of human language.

Amsterdam G.B.

Part IWhat is Linguistic Morphology?

This page intentionally left blank

1Morphology: basic notions

1.1 Relations between words

When you use an English dictionary to look up the diVerent meanings of

the verb walk, you will not be surprised that there are no separate entries

for walk, walks, and walked. You will also not feel disappointed if your

dictionary does not contain a separate entry for walking. If you come across

the sentence My staV walked out yesterday, and you want to Wnd out what

walked outmeans (‘‘go on strike’’) you will not look for an entrywalked out,

but rather for an entry walk out. In many dictionaries, walks, walked, and

walking are not even mentioned in the entry for walk. It is simply assumed

that the language user does not need this information. The reason for the

absence of this information is that these diVerent English words are felt to

be instantiations of the same word, for which walk is the citation form.

So we have to make a distinction between the notion ‘word’ in an abstract

sense (lexeme) and the notion ‘word’ in the sense of ‘concrete word as used

in a sentence’. The concrete words walk, walks, walked, and walking can be

qualiWed as word forms of the lexeme walk. Small capitals are used

to denote lexemes when necessary to avoid confusion between these two

notions ‘word’. English dictionaries assume that the language user will

be able to construct these diVerent forms of the lexeme walk by applying

1.1 Relations between words 3

1.2 Paradigmatic and syntagmatic

morphology 6

1.3 The functions of morphology 13

1.4 Morphology and the lexicon 16

1.5 The goals of morphology 23

Summary 24

Questions 24

Resources for morphology 26

Further reading 26

the relevant rules. These rules for computing the diVerent forms of lexemes

are called rules of inXection.

This example shows that dictionaries presuppose knowledge of relations

between words. It is the task of linguists to characterize the kind of

knowledge on which the awareness of the relation between the word

forms walk, walks, walked, and walking is based. Knowledge of a language

includes knowledge of the systematicity in the relationship between the

form and meaning of words. The words walk, walks, walked, and walking

show a relationship in form and meaning of a systematic nature, since

similar patterns occur for thousands of other verbs of English. The sub-

discipline of linguistics that deals with such patterns is called morphology.

The existence of such patterns also implies that words may have an internal

constituent structure. For instance, walking can be divided into the con-

stituents walk and -ing. Therefore, morphology deals with the internal

constituent structure of words as well.

Dictionary makers assume that these forms of the lexeme walk are

formed according to rules, and therefore need not be speciWed individually

in the dictionary. The same assumption plays a role in the case of nouns

and adjectives. For English nouns, the plural form does not need to be

speciWed in the dictionary if it is regular, and neither does the adverbial -ly

form in the case of adjectives. For example, my English–Dutch dictionary

(Martin and Tops 1984) does not mention the adverbs correctly and

economically in addition to correct and economical. On the other hand, it

does specify the adverb hardly. Why is that so? Is it due to inconsistency or

sloppiness on behalf of the dictionary makers, or is there a principled

reason behind this choice? There is indeed a reason: the meaning of hardly

cannot be predicted from that of hard and -ly.

This kind of knowledge is also relevant when searching for information

on the internet and in other digital data resources such as corpora of

actual language use and electronic dictionaries. Suppose you want to

collect information on tax. You might Wnd it helpful if the search engine

is programmed in such a way that it will not only recognize documents with

the word tax, but also documents with the words taxation, taxable, and

taxability as relevant. In fact, for many search engines this is not the

case. The words taxation and taxable are both derived from the verb to

tax which is related to the noun tax. The word taxability in its turn is

derived from taxable. Hence, we may qualify this set of related words as

a word family. On the other hand, when searching for information on tax

4 what is linguistic morphology?

issues, you would not like your search engine to retrieve documents with

the words taxi, taxis, taxon, or taxonomy that also begin with the letter

sequence tax. This example shows that analysis of the systematicity in the

relations between words is essential for the computational handling of

language data. What we need for this purpose is a morphological parser,

a computer program that decomposes words into relevant constituents:

tax-ation, tax-able, and tax-abil-ity.

There is an intuitive diVerence between the members of the word family

of tax mentioned above and the set of word forms walk, walks, walked,

walking. The diVerent words related to the verb to tax are not felt as forms

of the same word, but as diVerent though related words that each have their

own entry in the dictionary, that is, are diVerent lexemes. We speak here of

lexeme formation (or word-formation): taxability has been formed on the

basis of taxable through the addition of -ity, and taxable in its turn has

been formed on the basis of the verb tax, just like taxation. The verb tax

itself has been formed by turning the noun tax into a verb.

So far we have taken for granted that we can distinguish words from

other linguistic units such as phrases, and we are no doubt inXuenced by the

orthographical convention of using spaces to indicate word boundaries.

Determining if a particular linguistic unit is a word is not always that easy,

however, and certainly not for languages without a written tradition. Even

for English we might not be certain. Why is income tax to be considered as

a word rather than a phrase? After all, its constituents are separated by a

space in its spelt form. The issue of word demarcation is taken up a number

of times in this book.

Word-formation is traditionally divided into two kinds: derivation and

compounding. Whereas in compounding the constituents of a word are

themselves lexemes, this is not the case in derivation. For instance, -ity is

not a lexeme, and hence taxability is a case of derivation. The word

income tax, on the other hand, is a compound since both income and

tax are lexemes. Changing the word class of a word, as happened in the

creation of the verb to tax from the noun tax, is called conversion, and may

be subsumed under derivation.

Another dimension of this kind of knowledge about words assumed by

dictionary makers of English manifests itself in the fact that words that

are quite common in English might not be covered by a dictionary.

For instance, my English–Dutch dictionary does notmention bottle factory,

although it does mention bottle baby, bottle bank, bottleneck, and a number

morphology: basic notions 5

of other words beginning with bottle. Yet, I have no problem in under-

standing the title of the novel The Bottle Factory Outing written by Beryl

Bainbridge. What the dictionary presupposes is that the user of English

knows the words bottle and factory, and that the compound bottle factory

refers to a particular kind of factory, not to a particular kind of bottle: it is

the rightmost of the two word constituents that determines what kind

of thing the compound denotes. This is a systematic fact of English.

Therefore, one can understand the meaning of bottle factory without

having ever come across that word before. That also applies to the even

more complex word bottle factory outing. This example illustrates the

creative aspect of morphological knowledge: it enables us to understand

or coin new words. Morphological knowledge may thus lead to rule-

governed creativity in the use of language. If we want to be understood,

our new linguistic expressions must comply with the rules of the language.

It is these rules that enable every language user to produce and understand

linguistic expressions that she has never come across before.

The examples of morphological knowledge discussed so far come from

English. The reason for this choice in an introductory chapter is a practical

one: English is the language that all readers of this book are assumed to

understand. English is not the obvious choice when one wants to discuss

the nature of morphological systems in general, certainly not in the realm

of inXection. After all, English has a relatively poor inXectional system, in

which only a few grammatical distinctions are expressed. For instance,

whereas English has only four diVerent forms for regular verbs such as

walk, Romance languages such as French, Italian, and Spanish have tens

of diVerent forms for verbs. We should be aware of these considerable

diVerences in morphological richness between languages. Therefore, it is

important to look at a wide variety of languages in order to get a good idea

of the morphological possibilities of natural language.

1.2 Paradigmatic and syntagmatic morphology

The term ‘morphology’ has been taken over from biology where it is used

to denote the study of the forms of plants and animals. Its Wrst recorded

use is in writings by the German poet and writer Goethe in 1796. It was

Wrst used for linguistic purposes in 1859 by the German linguist August

Schleicher (Salmon 2000), to refer to the study of the form of words.


In present-day linguistics, the term ‘morphology’ refers to the study of the

internal structure of words, and of the systematic form–meaning correspond-

ences between words. Consider the following sets of English words:

(1) a. buy b. buyer

eat eater

paint painter

sell seller

send sender

In these sets of words we observe a systematic form–meaning correspond-

ence. The words in (1b) diVer from the words in (1a) in that they have an

additional part -er, and a corresponding meaning diVerence: each word in

(1b) has the meaning ‘‘one who Vs’’, where V stands for the meaning of the

corresponding verb in (1a). This is the basis for assigning a word such as

buyer an internal morphological constituency: buy-er. The form diVerences

between these two sets of words concern two properties: the words in (1b)

have the additional sound sequence [

e

r] (or [

e

] in standard British pronun-

ciation) compared to the words in (1a), and they are nouns, whereas the

words in (1a) are verbs. The form diVerences thus have a phonological and

a syntactic dimension. The meaning diVerence is quite clear: the nouns in

(1b) subsume the meaning of the corresponding verbs, and have some extra

meaning due to the presence of -er. Since the nouns are formally and

semantically more complex than the corresponding verbs, we will say that

the nouns have been derived from the verbs. That is, there is a direction

in the relationship between these two sets of words. The word buyer is

a complex word since it can be decomposed into the constituents buy and

-er. The word buy, on the other hand, is a simplex word, because it cannot

be decomposed any further into smaller meaningful units, only into sound

segments.

The notion ‘systematic’ in the deWnition of morphology given above is

important. For instance, we might observe a form diVerence and a corre-

sponding meaning diVerence between the English noun ear and the verb

hear. However, this pattern is not systematic: there are no similar word

pairs, and we cannot form new English verbs by adding h- to a noun. There

is no possible verb to heyewith the meaning ‘‘to see’’ derived from the noun

eye. Therefore, such pairs of words are of no relevance to morphology.

Similarly, we do not assign morphological constituency to German fressen

‘‘eating by animals’’ although it forms a pair with essen ‘‘to eat’’, since there


is no morphological constituent fr- that occurs in other word pairs as well.

The words fressen and essen are in fact related historically (fr- derives from

the early Germanic word fra), but fressen is no longer a complex word.

So words can lose their status of complex word.

The existence of related words with a systematic form–meaning diVerence

is crucial in assigning morphological structure to a word. The following

Dutch words for diVerent kinds of Wsh all end in -ing:

(2) bokking ‘‘bloater’’, haring ‘‘herring’’, paling ‘‘eel’’, wijting ‘‘whiting’’

Yet, we do not consider this -ing a morphological constituent with the

meaning ‘‘Wsh’’ because there are no corresponding Dutch words bok, haar,

paal, and wijt with a meaning related to the corresponding words ending in

-ing (these words do exist, but with a completely unrelated meaning).

The two sets of words given in (1) form paradigms. The term ‘paradigm’

is used here in a general sense to denote a set of linguistic elements with a

common property. All words in (1a) are verbs, and thus form a paradigm.

The same applies to the words in (1b) which are all nouns ending in -er. In

our deWnition of morphology as given above we see two diVerent perspec-

tives. When we speak about morphology as the study of the systematic

form–meaning correspondences between the words of a language, we take

a paradigmatic perspective, since we take properties of classes of words as

the starting point of morphological analysis. When morphology is deWned

as the study of the internal constituent structure of words, we take a

syntagmatic perspective.

We distinguish these two diVerent perspectives on language because lan-

guage units exhibit syntagmatic and paradigmatic relationships. They have a

syntagmatic relationship when they are combined into a larger linguistic

unit. For instance, the words the and book have a syntagmatic relationship

in the phrase the book. In contrast, the determiners a and the are para-

digmatically related: they belong to the set of determiners of English, and

can both occur at the beginning of a noun phrase, but never together: *the a

book. Hence, they belong to the paradigm of determiners of English.

A clear instantiation of a primarily syntagmatic approach to morph-

ology is morpheme-based morphology. In this approach, focus is on the

analysis of words into their constituent morphemes. That is, morphology is

conceived of as the syntax of morphemes, as the set of principles for com-

bining morphemes into words. Morphemes, the morphological building

blocks of words, are deWned as the minimal linguistic units with a lexical


or a grammatical meaning. For instance, the noun buyer consists of two

morphemes, buy and -er. The verbal morpheme buy is called a free or

lexical morpheme, because it can occur as a word by itself, whereas -er is

an aYx (hence a bound morpheme that cannot function as a word on its

own). This is indicated by the hyphen preceding this morpheme: it requires

another morpheme to appear before it in a word. Each of these morphemes

is listed in the morpheme list of English: eat as a morpheme of the category

Verb (V), and -er as an aYxal morpheme of the category Noun (N) that is

speciWed as occurring after verbs: [V —]. This speciWcation of the aYx -er

assigns it to the subcategory of aYxes that combine with verbs, and hence

we call it a subcategorization property of this aYx. The morphological

structure of eater might be represented as follows:

(3) [ [eat]V [er]N-aV ]N

This complex word can be created by the general mechanism of concaten-

ation, the combination of elements into a linear sequence. This word is well

formed because the requirement that -er occur after a verb is met. The fact

that this combination of morphemes is a noun, and not a verb, follows

from the generalization that English suYxes determine the category of

the complex words that they create: since -er is an aYxal noun, the whole

word is a noun.

Thus, the language user is able to coin new polymorphemic words (words

consisting of more than one morpheme) through the concatenation of

morphemes, and of morphemes with words that are themselves polymor-

phemic. An example of the latter is the formation of the verb tranquilize,

itself derived from tranquil through the addition of -ize. The formation of

tranquilizer is not a matter of concatenating three morphemes. Instead, it is

a two-steps operation. First, the bound morpheme -ize has been added

to the simplex adjective tranquil, resulting in the verb tranquilize. Sub-

sequently, the bound morpheme -er has been added to this verb. The

morphological structure of this word is therefore a layered one, and can

be represented in the form of a string with labelled bracketing, or as a tree

(Figure 1.1). In short, morphology might be seen as morpheme syntax,

as the set of principles that tell you how to combine free and bound

morphemes into well-formed words.

This syntagmatic approach can be contrasted to a primarily paradig-

matic approach to morphology. In the latter one, the creation of new

complex words is seen Wrst and foremost as the extension of a systematic


pattern of form–meaning relationships in a set of established words to new

cases, resulting in new words. Once we have discovered the abstract system-

atic pattern behind the words in (1), we will be able to extend this pattern to,

for instance, the verb swim, resulting in the word swimmer:

(4) Pattern [x]V : [x-er]N ‘‘one who Vs’’; swim : swimm-er

(the variable x stands for the string of sound segments of the verb). In the

gloss ‘‘one who Vs’’, the symbol V stands for the meaning of the verb. The

gloss indicates that nouns ending in -er have a meaning that encompasses

the meaning of the corresponding verb.

In this approach, it is not denied that the word swimmer consists of two

constituent morphemes, but they are not the basic building blocks. Instead,

words and relationships between words form the point of departure of the

morphological analysis, and morphemes have a secondary status in that

they Wgure as units of morphological analysis. Bound morphemes such as

-er do not have lexical entries of their own, and only exist as part of

complex words and of abstract morphological patterns such as (4).

From the point of view of language acquisition the paradigmatic per-

spective on complex words is the starting point of morphological analysis.

When acquiring one’s mother tongue, one has to discover the existence of

morphological patterns on the basis of the individual words encountered in

the input data. Only when language users have acquired a suYcient number

of words of the relevant type, can they conclude to a systematic abstract

pattern in sets of related words that might be used for the coinage of new

words.

This paradigmatic pattern can receive a syntagmatic interpretation as

well: the pattern can be interpreted as a morphological rule for the attach-

ment of bound morphemes to words. That is, paradigmatic relationships

[[tranquill ]A[ize]Vaff]V [er]Naff]N

Fig. 1.1 The morphological structure of tranquillizer


can be projected onto the syntagmatic axis of language structure. The

pattern in (4) can thus be interpreted as the following morphological rule:

(5) [x]V ! [ [x]V er]N ‘‘one who Vs’’

This rule states that nouns with a particular meaning (‘‘agent nouns’’) can

be derived from verbal lexemes by adding the aYx -er to the stem form of

the verb. On the left side of the arrow, the requirements on the properties

of the input words are speciWed, on the right side the formal and semantic

properties of the output words. The arrow indicates the direction of the

operation (input left, output right). The assumption of such aYx-speciWc

morphological rules means that bound morphemes do not exist as lexical

items of their own, but only as part of morphological rules. Consequently,

we get a slightly diVerent representation of the morphological structure

of the word tranquillizer given in Figure 1.2 (compare Figure 1.1).

Instead of assuming a rule for this word-formation pattern, one might

also express this regularity in the form of a template for the coining of new

nouns in -er of the following form that is formally equivalent to the

morphological rule (5):

(6) [ [x]V er]N ‘‘one who Vs’’

The morpheme concatenation approach and the lexeme-based rule

approach may in fact lead to similar analyses of word structure. In both

approaches, the polymorphemic noun swimmer will have the internal struc-

ture [ [swim]V er]N. The (minor) diVerence is that in the rule approach, the

bound morpheme -er has no lexical category label of its own, since it is not

a lexical entry. Yet, we should realize that rule (5), the rule interpretation of

an extendable word pattern, has a paradigmatic Xavour: it is not a rule

about morpheme concatenation, but it speciWes a formal and semantic

operation (aYx attachment and change of meaning) on lexemes. Similarly,

Fig. 1.2 The morphological structure of tranquillizer without aYx labels


template (6) is an abstract scheme that speciWes the common properties of a

set of words, but can also be used as a ‘recipe’ to create new words.

It is essential that morphological rules can take established words as their

inputs. If an established polymorphemic word has idiosyncratic properties,

these properties will recur in words derived from it. For example, the com-

plex noun transformation has a speciWc conventionalized meaning in

generative syntax (the change of a syntactic structure). Hence, the same

idiosyncratic meaning will recur in the adjective transformational derived

from this noun transformation. Similarly, the adjective edible not only

means that something can be eaten, but also that it can be eaten safely. This

idiosyncratic meaning aspect of edible recurs in the derived noun edibility.

Therefore, we must allow for established polymorphemic lexemes to func-

tion as the bases of word-formation. That is why morphological rules must

be lexeme-based.

A particular challenge for the morpheme-based approach to morphology

is the existence of morphological operations that do not consist of the

concatenations of morphemes, so called non-concatenative morphology.

The past tense forms of English irregular verbs, for instance, are not made

through addition of a morpheme to a stem, but by replacement of vowels,

as in sing-sang, and grow-grew. Another example (taken from Kutsch

Lojenga 1994: 135) is that Ngiti, a Central-Sudanic language of Congo

makes use of tones to distinguish morphologically related words. The

plural form of a number of nouns is made by replacing the tones of the

last two syllables (a sequence of a Mid tone and a Low tone) of the singular

noun by a High tone. (The grave and acute accents indicate Low and High

tones respectively; the absence of an accent indicates Mid tone.)

(7) singular plural

malimo malımo ‘‘teacher(s)’’

kama kama ‘‘chief(s)’’

malayika malayıka ‘‘angel(s)’’

This process of forming plural nouns cannot be stated straightforwardly in

a syntagmatic approach to morphology since there is no addition of a

(tonal) morpheme. This pattern can be expressed straightforwardly in

paradigmatic terms, as a systematic diVerence in form (tone pattern) correl-

ating with the semantic distinction between singular and plural. Such a

paradigmatic account of this regularity may look as in (8) where the

templates for singular and plural nouns of Ngiti are given:


(8) [ . . .VM:VL:]Nsg, [ . . .VH:VH:]Npl

The superscripts L, M, and H indicate the tone assigned to the vowel (V),

and the Vs stand for the last two vowels of the words.

There are also cases of paradigmatic word-formation, in which a new

word is formed by replacing one constituent with another. For instance, the

Dutch compound boeman ‘‘lit. boo-man’’ has a particular idiosyncratic

meaning ‘‘ogre, bugbear’’. Its female counterpart boevrouw has obviously

been coined by replacing the constituent man ‘‘man’’ with vrouw ‘‘woman’’

rather than by directly combining boe and vrouw into a compound, given

the fact that the two compounds share this idiosyncratic meaning. Such a

case of word-formation cannot be accounted for in a purely syntagmatic

approach to morphology, neither a morpheme-based nor a rule-based one.

It is based on speciWc words, and therefore a typical case of analogy:

(9) man : vrouw ¼ boeman : boevrouw ‘‘female bugbear’’

The paradigmatically oriented deWnition of morphology given above

expresses directly that morphology is lexeme-based. Lexemes form the

point of departure of morphological processes. In lexeme formation (or

word-formation) we create new lexemes on the basis of other lexemes,

whereas in inXection, speciWc forms of lexemes are computed (instead of

lexeme formation we will speak of word-formation when there is no

risk of misunderstanding). The processes of word-formation and inXection

together form the morphological part of a grammar.

Morphology deals with both the form and the meaning of linguistic

expressions. Hence, one might qualify morphology as word grammar,

that part of the grammar that accounts for the systematic form–meaning

relations between words. In other words, it is a set of correspondence rules

between forms and meanings of words. The notion ‘word grammar’ stands

in opposition to ‘sentence grammar’, the grammar which describes the

systematic relations between form and meaning at the sentence level.

1.3 The functions of morphology

The two basic functions of morphological operations are (i) the creation of

new words (i.e. new lexemes), and (ii) spelling out the appropriate form of a

lexeme in a particular syntactic context.


An example of the Wrst function, lexeme formation, is given in section

1.1: the coining of the word bottle factory from the existing lexemes bottle

and factory. Morphology thus provides means for extending the set of

words of a language in a systematic way. The coinage of bottle factory is

a case of compounding, in which two lexemes are combined into a new one.

In the other type of word-formation, derivation, exempliWed by the word

swimmer, use is made of morphological operations on lexemes, whereas in

compounding, two or more lexemes are combined into a new word.

Why do we need new words? One obvious reason is that language users

need new expressions for new objects, or for new concepts. Once there is an

entity or concept ‘‘factory for the production of bottles’’, it is quite easy to be

able to refer to such a concept with one word, bottle factory instead of using a

circumscription. Thus, word-formation has a labelling function. Creating

a word label for a new kind of entity, event, or property may have the

additional pragmatic advantage that it draws attention to the new concept

involved. For instance, the word construction grammar has been created to

denote a particular school of linguistic thought in which the linguistic notion

‘construction’ plays a central role. By coining this label, a new linguistic

school has been established, and thus its ideas will draw attention more

easily. New verbs have been created to express new types of events or actions,

such as the English verbs in -ize: legal-ize ‘‘to make legal’’, tranquil-ize ‘‘to

make tranquil’’, that express the causation of an event or property.

However, this is not the only function of word-formation. Another

important function is that of syntactic recategorization: by using morpho-

logically related words of diVerent syntactic categories, we achieve stylistic

variation and text cohesion, as the following examples (from Kastovsky

1986: 595) show:

(10) He made Wsts . . . He deWsted to gesture.

If that’s not civil, civilize it, and tell me.

[ . . . ] and whether our own conversation doesn’t sound a little potty. It’s the

pottiness, you know, that is so awful.

A pragmatic reason for coining new words is found in the domain of

evaluative morphology. In many languages diminutive forms of words are

not used primarily for indicating the small size of the object denoted, but

for giving a positive or negative evaluation. For instance, the Portuguese

diminutive noun avozinho (from avo ‘‘grandfather’’) means ‘‘dear grand-

father’’ rather than ‘‘small grandfather’’, and in Dutch the diminutive noun


baantje ‘‘job’’ derived from baan ‘‘job’’ is used to refer to a job without

prestige. A related phenomenon is that of the use of attenuative forms.

The English morpheme -ish is often used to express the notion ‘‘sort of,

not exactly’’: when we use nine-ish instead of nine as the time for an

appointment, we mean that we do not expect people to be there at nine

sharp. Thus, we might use morphology to express our subjective feelings

towards something or to weaken or relativize a notion.

The function of inXection is primarily that of making forms of lexemes,

including the correct forms of a lexeme appropriate for particular contexts.

For instance, in an English clause, the verb has to agree with the subject

with respect to number (singular or plural) and person (third or non-third),

and this determines the choice between walk and walks: in a clause with

present tense, walks has to be chosen if the subject is third person singular,

and walk otherwise. In many languages, the form of a noun is determined

by its syntactic context, and each noun has a number of cases. For instance,

the Polish noun kot ‘‘cat’’ has the case forms shown in (11). We call this

structured set of word forms the inXectional paradigm of this lexeme (note

that this is a more speciWc use of the notion ‘paradigm’ as introduced above

in section 1.2). The term ‘inXectional paradigm’ may also be used to denote

the abstract inXectional pattern, the set of labelled cells that these word

forms occupy. As can be read oV this paradigm of case forms, when the

lexeme kot occurs in direct object position and therefore has accusative

case, the word form kota has to be used if the word has a singular meaning,

and the form koty if it has a plural meaning. That is, one of the accusative

forms has to be chosen for this syntactic position.

(11) singular plural

nominative kot kot-y ‘‘cat, subject’’

genitive kot-a kot-ow ‘‘of the cat’’

dative kot-u kot-om ‘‘to the cat’’

accusative kot-a kot-y ‘‘cat, object’’

instrumental kot-em kot-ami ‘‘with the cat’’

locative koci-e kot-ach ‘‘on the cat’’

vocative koci-e kot-y ‘‘o, cat’’

Another function of morphology is that the relation between sentences

in a text can be established by using morphological markers of coreferenti-

ality. In Wambon, a language of New Guinea (examples from de Vries


1989: 62), verbal forms have Same Subject (SS) forms and DiVerent Subject

(DS) forms (1sg ¼ Wrst person singular, 3sg ¼ third person singular, nf ¼non-future):

(12) Nukhe oye khetak-mbel-o topkeka-lepo

I pig see-ss-coord Xee-1sg.past

‘‘I saw a pig and I Xed’’

(13) Nukhe oye khetakha-lev-o topkeka-tmbo

I pig see-1sg.nf.ds-coord Xee-3sg.past

‘‘I saw a pig and it Xed’’

Both examples consist of two clauses, with the coordinating element -o

linking these two clauses. This coordinative morpheme is attached to the

verb of the Wrst clause. The word khetakmbelo in (12) has the Same Subject

form, which indicates that in both clauses we have the same subject ‘‘I’’.

In (13), on the other hand, the word khetakhalevo is a DiVerent Subject

form, which indicates that the subject of the next clause is a diVerent one.

It is not ‘‘I’’, but the pig that Xed. This kind of subject marking is called

switch reference.

1.4 Morphology and the lexicon

The set of lexemes of a language comprises two subsets: simplex lexemes and

complex lexemes. These lexemes are listed in the lexicon to the extent that

they are established, conventionalized units. A complex lexeme likeninish is

a well-formed lexeme of English, but need not be listed in the lexicon since it

is completely regular, and there is no conventionalization involved.

The lexicon speciWes the properties of each word, its phonological form, its

morphological and syntactic properties, and its meaning. The basic structure

of lexical entries for the lexemes swim and swimmermay look as follows:

(14) /swØm/ /swØm

e

r/

[x]V [[x]V er]N

swimactivity person performing swimactivity

The Wrst line in these lexical entries speciWes the phonological form of these

lexemes: a sequence of sound segments between slashes. On the second line,

categorial information, and internal morphological structure of a word are

speciWed. On the third line, the meaning of the lexeme is speciWed, here


indicated by the use of small capital letters. The subscript activity speciWes

the type of event expressed by this verb. A lexical entry thus expresses a

correspondence between phonological, syntactic, and semantic pieces of

information, just like morphological rules or templates, which do the same

at a more abstract level, in a generalized fashion, with variables taking the

place of the individual properties of lexemes.

Most complex words have been derived by one of the available word-

formation processes of a language. Indeed, as we saw above, one of the

main functions of morphology is to expand the set of available words.

Once a complex word has been formed, it may get established as a word of

the language. This means that it is used by more than one native speaker,

and on diVerent occasions, and that language users will recognize it as a

word they have come across before. The set of established words of

a language functions as the lexical norm or lexical convention of that

language. For instance, in British English the machine that is used for

drawing money from one’s bank account is called a cash dispenser, and in

American English it is called an automatic teller machine (ATM). In fact, it

would also have been possible to use the compoundmoney machine for this

device, but the established words function as a lexical norm, and hence they

can block the creation of the compoundmoney machine. That is, the lexicon

as the set of established lexical units of a language may have a blocking

eVect on the creation of new words. It does not mean that money machine

is an ill-formed word, only that its use might not be appropriate.

When a possible word has become an established word, we say that it has

lexicalized. An important eVect of lexicalization of complex words is that one

of its constituent words may get lost, whereas the complex word survives.

For instance, the Dutch verb vergeet ‘‘to forget’’ no longer has a simplex

counterpart geet, unlike its English counterpart forget for which the corre-

sponding word get does exist. We therefore consider vergeet a formally

complexword. It still behaves as a complex verb since it selects a past participle

without thepreWxge-, just likeotherpreWxedverbsofDutch.For example, the

past participle of the preWxed verb ver-wacht ‘‘to expect’’ (derived from

the verb wacht ‘‘to wait’’) is verwacht. Similarly, the past participle of the

verb vergeet is vergeten, not *gevergeten. This may be contrasted with the

verb verbaliseer ‘‘to Wne’’ in which the part ver- has no preWx status.

The past participle of this verb is ge-verbaliseer-d, with the preWx ge- present.

The term ‘lexicalization’ is also used for a related phenomenon, namely

that established words may have idiosyncratic, unpredictable, properties.


The meaning of honeymoon, for example, is not predictable on the basis of

the meanings of its constituent lexemes honey and moon, and this requires

this compound to be listed in the lexicon. Having idiosyncratic properties

thus implies for a word that it has to be listed, but the inverse is not

necessarily true: a complex word that is listed may have fully predictable

properties, and may be listed only because it is an established word, that is,

belongs to the lexical norm.

The notion ‘lexicon’ refers to the repository of all information concerning

the established words and other established expressions of a language. It is an

abstract linguistic entity, tobedistinguished fromthenotion dictionary, which

refers to practical sources of lexical information for the language user in some

material (paper or electronic) form. A dictionary will never provide a full

coverage of the lexicon due to practical limitations of size and requirements

of user-friendliness, and because the lexicon is expanding and changing

daily. The third related notion of relevance here is that of themental lexicon,

the mental representation of lexical knowledge in the brain of the individual

language user. The mental lexicon of an individual is always smaller than the

lexicon in the linguistic sense: nobody knows all the established words of a

language. Moreover, the mental lexicon exhibits an asymmetry between

production and perception: we understand probably about Wve times more

words of our mother tongue than we actually use in language production.

In many languages, morphology is extremely important for the size of the

lexicon. In all European languages, the number of established complex words

is much higher than the number of simplex words. Consequently, the mor-

phological rules of a language have two functions: they indicate how new

lexemes and word forms can be made, and they function as redundancy rules

with respect to the established complexwords of a language. For instance, the

lexical information that lover is a noun, and that the meaning of this word

comprises that of the verb love is redundant information. These properties

are speciWed in rule (5).On theother hand, the information that this noun is an

established word of English, with a particular idiosyncratic meaning ‘‘male

sweetheart, suitor’’ is unpredictable, non-redundant lexical information.

Morphological patterns that can be systematically extended are called

productive. The derivation of nouns ending in -er from verbs is productive

in English, but the derivation of nouns in -th from adjectives is not: it is

hard to expand the set of words of this type such as depth, health, length,

strength, and wealth. Marchand (1969: 349) has observed some occasional

coinings like coolth (after warmth), but notes that such word coinings are


often jocular, and hence do not represent a productive pattern. If we want

to coin a new English noun on the basis of an adjective, we have to use -ness

or -ity instead. In the case of unproductive patterns, the morphological

rule involved functions as a redundancy rule only, and not as a rule for

the creation of new words.

Lexical storage of complex morphological forms is also relevant in

the realm of inXection. For example, Dutch has two plural endings for

nouns, -s and -en. The second one is normally used for words consisting of

one syllable. In the case of the monosyllabic noun boon ‘‘bean’’, the regular

plural is bon-en, as expected. However, for zoon ‘‘son’’ both the irregular

zoon-s and the regular zon-en can be used. Hence, the plural form zoon-s

has to be speciWed in the lexicon.

The morphological system of a language is not its only source of complex

words. There are at least three other sources: borrowing, phrases becoming

words, and word creation.

As to borrowing, European languages have borrowed many words from

Greek and Latin, often with French as the intermediary language. Consider

the following list of Dutch verbs and their English glosses:

(15) deduceer ‘‘deduce’’

induceer ‘‘induce’’

produceer ‘‘produce’’

reduceer ‘‘reduce’’

reproduceer ‘‘reproduce’’

A verb like produceer can be analysed into three parts: pro-duc-eer, that is,

it is a polymorphemic word. The constituent -eer is a recurrent part of all

these words, and so is -duc-. The sequences de-, in-, pro-, and re- are also

recognizable elements in this set of verbs. Yet, we cannot say that these

verbs have been created by a rule of Dutch or English morphology since

there is no lexeme duc from which these words could have been derived.

Instead, a word such as produceer has been created by transforming the

originally Latin verb producere, and by adapting its form by turning the

ending -ere into -eer. The polymorphemic nature of such words remains

recognizable in the borrowing languages. These borrowing patterns have

led to a pan-European lexicon, a large stock of cognate complex words in

the major languages of Europe.

A second non-morphological source of complex words is the univerba-

tion (‘‘becoming a word’’) of phrases. Phrases may lexicalize into words,


and thus lead to complex words. Examples from English are jack-in-the-

box, forget-me-not (nouns), and dyed-in-the-wool, down-at-heel, over-the-top

(adjectives). The following Dutch words all begin with te-, originally a

preposition, the etymological cognate of English to:

(16) te-gelijker-tijd ‘‘lit. at same time, simultaneously’’

te-rug ‘‘lit. to back, back’’

te-vreden ‘‘lit. at peace, satisWed’’

te-zamen ‘‘together’’

In the Wrst example, tegelijkertijd, the three constituents are clearly recog-

nizable, and theirmeanings are relevant. Thewords gelijk ‘‘identical, same’’,

and tijd ‘‘time’’ are current words of Dutch (the form of gelijk used here is

gelijker, with an old inXectional ending -er). Therefore, tegelijkertijd is

a complex, polymorphemic word. So the fact that a word is polymorphemic

does not imply that it has been created by morphological rule. The second

example, the word terug, is also interesting because it serves to illustrate

a recurring problem of analysis for the linguist: when do we consider a word

complex? Although rug ‘‘back’’ is a word of Dutch, it remains to be seen if

we should consider terug a simplex or a complex word. In fact, many native

speakers do not recognize the word rug in terug because of themore abstract

meaning of terug, which no longer refers to a part of the human body.

Language users may also make new words by means of word creation (or

word manufacturing). The following types can be distinguished:

(17) blends: combinations of the Wrst part of one word with the second part of

another: brunch < breakfast þ lunch; stagXation < stagnation þ inXation;

acronyms: combination of initial letters of a word sequence: NATO < North

Atlantic Treaty Organization; French OTAN < Organisation du Traite

de l’Atlantique Nord

alphabetisms: combination of the Wrst letters of words, pronounced with the

phonetic value of these letters in the alphabet: French SVP < S’ıl vous plaıt

‘‘please’’; Dutch KLM < Koninklijke Luchtvaart Maatschappij ‘‘royal airline

company’’; English CD ‘‘compact disc’’, SMS ‘‘Short Message Service’’;

clippings: one or more syllables of a word: mike < microphone, demo

< demonstration, French labo < laboratoire ‘‘laboratory’’, German Uni

< Universitat ‘‘university’’.

In the case of compounds, only one of them may be shortened, as

in GermanU-Bahn <Untergrund-bahn ‘‘metro’’, English e-mail ‘‘electronic

mail’’, and FAQ-list ‘‘frequently asked questions list’’. In ellipsis, the Wrst


constituent is taken to represent the whole as in Dutch VU < VU-Ziekenhuis

‘‘Free University Hospital’’ (VU is itself an acronym for Vrije Universiteit

‘‘Free University’’). The diVerence between the hospital and the university

meaning of the acronym can still be expressed, because these words diVer

in gender, and hence select diVerent deWnite articles: het VU (hospital) versus

de VU (university).

Instead of the term clipping, linguists also use the term truncation, espe-

cially in relation to the formation of personal names which have an aVective

load and function as hypocoristics (names of endearment). In many cases,

the stressed syllable of the full form is the core of the truncated name, which

consists of one or two syllables (the acute accent indicates word stress):

(18) English Dave < David, Liz < Elızabeth, Kate < Katherine, Sue < Susan

Dutch Hans < Johannes, Henk < Hendrik, Sanne < Suzanne

French Dom < Dominıque, Val < Valerıe, Fab < Fabrıce

Spanish Dina < Alexandrına, Marga < Margarıta, Neto < Ernesto

Truncation may operate in connection with the addition of an ending. In

English the endings -y or -ie can be added to the truncated form, in German

these truncated names may end in -i, -e, or -o:

(19) English Becky < Rebecca, Suzy < Suzanne, commie < communist

German Andi < Andreas, Daggi < Dagmar, Fundi < Fundamentalist,

Wolle < Wolfgang, Realo < Realist

A characteristic of word creation is that it makes use of reduction for the

creation of new words, unlike normal morphology. Consequently, the

meaning of the new word cannot be derived from its form straightfor-

wardly, and it therefore lacks semantic transparency. In blending, for

instance, the constituent parts that determine the meaning are not fully

present in the word, and hence the meaning is not recoverable from these

constituents. In stagXation, the parts stag- and -Xation do not have them-

selves the meanings ‘‘stagnation’’ and ‘‘inXation’’ respectively. Thus, the

English language user cannot know what the word stagXation means when

hearing or reading it for the Wrst time, unlike what is the case when one

encounters bottle factory for the Wrst time. This also applies to acronyms: if

you do not know a certain acronym, there is no way to Wnd out about its

meaning on the basis of your knowledge of the language. In the case of

clipping, its full form is not recoverable on the basis of the clipping, and so

its meaning is also unpredictable, although it is sometimes possible to


guess. Word creation is thus diVerent from word-formation in the strictly

morphological sense, where the meaning of the newly created word is

recoverable from that of its constituents, and it is typically an intentional

form of language use. The lack of transparency of these words serves to

create incrowd groups who understand these shortened words, and so they

have an important sociolinguistic value. They may also establish intimacy

(as is the case for truncated personal names), or informality (GermanUni is

more informal than Universitat ‘‘university’’).

This overview of word creation does not exhaust the set of special form–

meaning correspondences in words. There is also echo-word-formation, a

kind of reduplication, as in English zigzag, chitchat, French fou-fou ‘‘some-

what mad’’ or with rhyming words (Dutch ietsiepietsie ‘‘a little bit’’,

ukkepuk ‘‘small child’’), and sound symbolism in words beginning with the

same sound sequence. For instance, words with initial sw- typically denote

swinging movements (sweep, swing, swingle, etc.), and the following Dutch

words with kr- all refer to unpleasant, twisted notions: krijs ‘‘to shout’’,

kramp ‘‘cramp’’, krank ‘‘ill’’, krimp ‘‘shrink’’.

The similarity at a more abstract level between morphology proper and

word creation is that both are based on patterns of paradigmatic relation-

ships between sets of words.

So far, we have seen that the set of established words of a language can

be expanded in a number of ways. The lexicon, however, is not just a set of

words, but also comprises word combinations. For example, English (like

most Germanic languages) has many verb–particle combinations, also

called phrasal verbs of the type to look up which clearly consist of two

words which are even separable:

(20) a. The student looked up the information

b. The student looked the information up

The verb look up cannot be one word since its two parts can be separated, as

in sentence (20b). A basic assumption in morphology is the hypothesis of

Lexical Integrity: the constituents of a complex word cannot be operated

upon by syntactic rules. Put diVerently: words behave as atoms with respect

to syntactic rules, which cannot look inside the word and see its internal

morphological structure. Hence, the movement of up to the end of the

sentence in (20b) can only be accounted for if look up is a combination of

two words. That is, phrasal verbs such as look up are certainly lexical units,

but not words. Words are just a subset of the lexical units of a language.


Another way of putting this is to say that look up is a listeme but not a

lexeme of English (DiSciullo and Williams 1987).

Other examples of lexical multi-word units are adjective noun combin-

ations such as red tape, big toe, atomic bomb, and industrial output. Such

phrases are established terms for referring to certain kinds of entities, and

hence they must be listed in the lexicon. Some languages tend to prefer such

multi-word units to morphologically complex words as denoting expres-

sions. This is for instance the case for the Papuan languages of NewGuinea.

In sum,morphology is only one of themeans for expanding the lexicon of a

language: there are other ways of creating lexical units, and the set of complex

words can be enlarged by other means than regular word-formation.

1.5 The goals of morphology

The wordmorphology can be used in two ways: it refers to a subdiscipline of

linguistics, but it may also be used to refer to that part of the grammar of a

language that contains the rules for inXection and word-formation, that is,

the word grammar. This kind of ambiguity also applies to words like phon-

ology, syntax, and semantics. When we talk about the goals of morphology,

it is obviously the Wrst meaning of the word that is relevant here.

Why do linguists want to do morphology? The Wrst reason is that it is the

linguists’ task to describe and analyse the languages of the world as

accurately and as insightfully as possible. Hence, they have to deal with

the morphological phenomena of a language, and therefore need a set of

tools for description. Morphology provides such tools, a set of analytic

notions, which are discussed in greater detail in Chapter 2. A related,

second goal of linguists is developing a typology of languages: what are

the dimensions along which languages diVer, and how are these dimensions

of variation related and restricted? Do all languages have morphology,

and of all possible kinds? Are there explanations for the morphological

similarities and diVerences between languages? The kinds of morphology

that we come across in the languages of the world are discussed in greater

detail in Part II (word-formation) and Part III (inXection).

Third, morphology is a probe into the nature of linguistic systems,

and hence into human, natural language. For example, morphology quite

clearly shows that linguistic structure has two axes, a syntagmatic axis and

a paradigmatic one. Morphology also serves to get a better understanding


of the nature of linguistic rules and the internal organization of the

grammar of natural languages. Thus we may get to know more about

the architecture of the human language faculty and about the nature of

rule-governed creativity in the domain of language (Part IV).

Finally, morphology can be used to get a better insight as to how lin-

guistic rules function in language perception and production, and how

linguistic knowledge is mentally represented. Both psychological and

historical evidence throw light on this issue. Thus, morphology contributes

to the wider goals of cognitive science that explores the cognitive abilities of

human beings (Part V).

Summary

Morphology, the study of the internal structure of words, deals with the

forms of lexemes (inXection), and with the ways in which lexemes are

formed (word-formation). New words are made on the basis of patterns

of form-meaning correspondence between existing words. Paradigmatic

relationships between words are therefore essential, and morphology can-

not be conceived of as ‘the syntax of morphemes’ or ‘syntax below the word

level’. Morphology serves to expand the lexicon, the set of established

words of a language, but is not the only source of lexical units, and not

even that of all complex words, which also arise through borrowing,

univerbation, and word creation.

The established (simplex and complex) words of a language are listed in

the lexicon, an abstract linguistic notion, to be distinguished from the notions

‘dictionary’ and ‘mental lexicon’. Morphological rules have two functions:

they specify the predictable properties of the complex words listed in

the lexicon, and indicate how new words and word forms can be made.

Morphology as a subdiscipline of linguistics aims at adequate language

description, at the development of a proper language typology, and at

contributing to debates on the organization of grammars and the mental

representation of linguistic competence.

Questions

1. My English–Dutch dictionary does not mention all English adverbs in -ly.However, it does mention the adverb supposedly. Why do you think thisexception is made?


2. Why should we consider taxable a case of derivation rather than com-pounding although able is a lexeme of English?

3. Give the morphological structure (labelled bracketing) of the followingEnglish words: unhappiness, contrastive, disconnecting, contradiction,blue-eyed, connectivity.

4. Try to determine the morphological constituency (if any) of the followingEnglish words: colleague, cordial, correlate, electrometer, elongation,evaporate, eternity, euphemism, habitual, happy, music, negotiable, perform-ance, theology. To what extent do these words raise problems for morpho-logical analysis?

5. Here is a set of pairs of singular and plural nouns in Oromo, a languagespoken in Ethiopia and Kenya (Stroomer 1987:76–7).

raadda raaddoollee ‘‘young cow(s)’’uwaa uwoollee ‘‘woman/women’’eela eeloota ‘‘well(s)’’kobee kobeellee ‘‘shoe(s)’’kobee koboota ‘‘shoe(s)’’harree harreellee ‘‘donkey(s)’’sangaa sangoollee ‘‘ox/oxen’’

Which plural endings are found in these data?

6. Consider the following English verbs: forbid, forget, forgive, forgo, forswear.What evidence can you adduce for these verbs being complex?

7. What kind of sound symbolism may be involved in the following Englishverbs: spew, spit, spout, spatter, sprout, sprawl ?

8. A bilingual child with parents who have different native languages may besaid to have a mother tongue and a father tongue. How can we account forthe coining of father tongue?

9. Blending is quite popular as a means of creating new English words. Try tocome up with some meaning for the following recent blends: falloween,giraffiti, metrosexual, nicotini, pedlock. Why can’t you be sure about theirmeanings if you happen not to know them? (You can find the meanings ofthese words on the website www.wordspy.com.)

10. Bacronyms are words that are reinterpreted as acronyms, for instanceGeorge as a name for an organization with the acronymic interpretationGeorgetown Environmentalists Organized against Rats, Garbage, andEmissions.

a. Which kind of word creation is the word bacronym itself an instanti-ation of ?

b. Do you know bacronyms in your native language?


www.wordspy.com

Resources for morphology

The internet is a very useful resource of information about languages and

linguistics. Important websites for morphologists are Linguist List

(www.linguistlist.org), the website of the Summer Institute of Linguistics

(www.sil.org), with many links to other relevant websites, and www.

yourdictionary.com where one can Wnd morphological subgrammars

and morphological parsers for a number of languages. A survey of the

languages of the world is found in Grimes (2003): www.ethnologue.com.

A website devoted to recent English neologisms is www.wordspy.com.

Two recent handbooks of morphology are Spencer and Zwicky (1998),

and Booij et al. (2000–4 ¼ BLM in the references; in vol. i you will Wnd a

number of articles on the history of morphological research). The classical

handbook of English morphology is Marchand (1969). Spencer (1991)

and Carstairs-McCarthy (1992) are textbooks on morphology, with a lot

of attention to current theoretical debates in morphology. A number of

important articles on morphology have been reprinted in Katamba (2004).

Further reading

Morpheme-based morphology is defended in Lieber (1980), Selkirk (1982),and DiSciullo and Williams (1987). Morphology as ‘syntax below the wordlevel’ is argued for in Lieber (1992).The notion word-based (¼ lexeme-based) morphology is defended in Aron-

off (1976) and in Anderson (1992). The importance of paradigmatic relation-ships for morphology is highlighted in van Marle (1985) and Becker (1990a).The relation between morphology and the lexicon is discussed in detail in a

number of publications by Jackendoff (1975, 1997, 2002). The principle ofLexical Integrity is defended in Bresnan and Mchombo (1995).The functions of word creation are clarified in Ronneberger-Sibold (2000).

Word creation and sound symbolism in English are dealt with extensively inMarchand (1969).


www.linguistlist.org

www.sil.org

www.yourdictionary.com

www.yourdictionary.com

www.ethnologue.com

www.wordspy.com

2Morphological analysis

2.1 The atoms of words

Words can be chopped into smaller pieces. At the phonological level, words

can be divided into syllables or segments, and segments into their constitu-

ent phonological features. At the morphological level, words may consist

of more than one unit as well, which we may call the morphological atoms

of a word: pieces that are no further divisible into morphological subparts.

Just as there are diVerent kinds of atom in chemistry, there are diVerent

kinds of atom in morphology, and it is quite useful for morphological

analysis to be acquainted with their classiWcation. A good classiWcation is

an important analytic instrument, developed in order to get a better under-

standing of the structure and formation of words.

As we saw in Chapter 1, the Polish lexeme kot ‘‘cat’’ has a paradigm of

case forms; compare this to the case forms of the noun kobieta ‘‘woman’’ in

(1). Each cell of the paradigm of Polish nouns is occupied by a grammatical

(1) singular plural

nominative kot kobiet-a kot-y kobiet-y

genitive kot-a kobiet-y kot-ow kobiet

dative kot-u kobieci-e kot-om kobiet-om

accusative kot-a kobiet-e kot-y kobiet-y

instrumental kot-em kobiet-a kot-ami kobiet-ami

locative koci-e kobieci-e kot-ach kobiet-ach

vocative koci-e kobiet-o kot-y kobiet-y

2.1 The atoms of words 27

2.2 Morphological operations 34

2.3 Morphological typology 41

Summary 45

Questions 45

Further reading 47

word, i.e. a form of a lexeme with a particular property for the grammatical

categories number and case. Grammatical words may share the same word

form. For instance, both the gen.sg and the acc.sg form of kot have the

form kot-a. The phenomenon that two or more grammatical words have

the same word form is called syncretism. The distinction between lexeme,

grammatical word, and word form shows that the general notion ‘word’

subsumes a number of diVerent notions. In most cases it is clear which

interpretation of ‘word’ is intended, but sometimes it will be necessary to

use the more speciWc notions.

Each of the word forms of kot consists of a stem and an inXectional

ending (or desinence). The stem of a word is the word form minus its

inXectional aYxes, in this example kot-. It is the stem that forms the basis

for word-formation, not the whole word form. This might not be so clear

for the Polish noun kot, because the nom.sg word form kot of this word

happens to have no overt ending. However, the noun kobieta does have an

overt ending. For that reason, one may speak of a zero-ending for

the nom.sg. form of kot, and likewise for the gen.pl form of kobieta.

The following example from Italian also illustrates the role of the stem. The

singular form of macchina ‘‘machine’’ has the inXectional ending -a, and

the plural ending is -e:

(2) macchin-a ‘‘machine’’ macchin-e ‘‘machines’’ macchin-ista ‘‘machinist’’

It is the stem macchin- that is used as the basis for word-formation, as

shown by macchinista. In English, the form of the stem is identical to that

of the sg word form, and this is why English morphology is sometimes

qualiWed as word-based morphology, in contrast to the stem-based morph-

ology of, for instance, most Romance and Slavic languages. This is a

superWcial diVerence: these languages all have lexeme-based morphology,

they only diVer in that the stem-forms of lexemes do not always corres-

pond to word forms.

Stems can be either simplex or complex. If they are simplex they are

called roots. Rootsmay be turned into stems by the addition of amorpheme,

as the following examples from Polish (Szymanek 1989: 87) illustrate:

(3) a. butelk-a ‘‘bottle’’ b. butelk-owa-c ‘‘to bottle’’

Wltr ‘‘Wlter’’ Wltr-owa-c ‘‘to Wlter’’

bial-y ‘‘white’’ biel-i-c ‘‘to whiten’’

głuch-y ‘‘deaf’’ głuch-na-c ‘‘to become deaf ’’


The verbs in (3b) are given here in their citation form, the inWnitive. The

citation form is the form in which a word is mentioned when we talk about

it, and the form in which it is listed in a dictionary. In many languages, the

inWnitive is the citation form of a verb. In languages with case, the nom.sg

form is the citation form of nouns. Each of these Polish inWnitives consists

of a root, followed by a verbalizing morpheme that turns the root into a

stem, and is followed by the inWnitival ending -c. It is the stem-forms that

are used when new words are derived from these verbs.

Stem-forming suYxes play an important role in many Indo-European

languages. Italian verbs, for instance, have a thematic vowel after the

root morpheme, and this thematic vowel recurs in words derived from

these verbs:

(4) larg-o ‘‘wide’’ al-larg-a-re ‘‘to widen’’

profond-o ‘‘deep’’ ap-profond-i-re ‘‘to deepen’’

al-larg-a-ment-o ‘‘widening’’

ap-profond-i-ment-o ‘‘deepening’’

The thematic vowel is not a part of the root, as it does not occur in the roots

larg- and profond-. On the other hand, it cannot be seen as part of the

inWnitival suYx, because we do not want to miss the generalization that all

inWnitives end in -re. Hence, the vowels preceding the ending -re must be

assigned a morphological status of their own. Consequently, the noun

allargamento contains Wve morphemes: a preWx al-, a root larg, a thematic

vowel -a-, the derivational morpheme -ment, and the inXectional ending -o.

So this word has Wve morphological atoms, which cannot be decomposed

further into smaller morphological constituents. Each of these Wve atoms

has a diVerent name because they have diVerent functions in the make-up

of this word.

The general term for boundmorphemes that are added to roots and stems

is aYx. If an aYx appears before the root/stem, it is a preWx, if it appears

after the root/stem, it is a suYx. So al- and ap- are preWxes, whereas -a,

-ment, and -o are suYxes. Two other types of aYxation are illustrated in (5):

(5) inWx (within a root): Khmu (Laos) s-m-ka:t ‘‘roughen’’ < ska:t ‘‘rough’’;

Alabama (Stump 2001: 131) ho-chi-fna ‘‘smell, 2sg’’ < hofna ‘‘to smell’’, chiWp-

as-ka ‘‘poke, 2pl’’ < chiWpka ‘‘to poke’’;

circumWx (combination of preWx and suYx): Dutch ge-Wets-t ‘‘cycled, past

participle’’ < Wets ‘‘to cycle’’; German Ge-sing-e ‘‘singing’’ < sing ‘‘to sing’’

morphological analysis 29

InWxation and circumWxation are much rarer than preWxation and

suYxation.

AYxes are bound morphemes, but not all bound morphemes are aYxes.

There are many roots from Greek and Latin that are used in so called

neo-classical compounds but do not occur as words by themselves. These

compounds are called ‘neo-classical’ because they consist of constituents

from the classical languages Greek and Latin that were combined into

compounds long after these languages ceased to be ‘living languages’.

In such compounds either one or both constituents are not lexemes:

(6) micro-: micro-scope, micro-phone, micro-gram, micro-wave

tele-: tele-phone, tele-vision, tele-communication

-graph: di-graph, sono-graph, photo-graph, tele-graph

-scope: micro-scope, tele-scope, cine-scope, spectro-scope

Neo-classical roots such as scope and graph can also be used nowadays as

words, but in that case they have a more speciWc meaning than in these

compounds. Such non-lexical roots are called combining forms since they

only occur in combinationwith other morphemes. These bound roots cannot

be considered aYxes since that would imply that words such as necrology

would consist of aYxes only. This goes against the idea that each word has

at least one stem. Thus, we might adapt our deWnition of what compounds

are, and deWne them as combinations of lexemes and/or non-aYxal roots.

The bound morphemes in neo-classical compounds have an identiWable

meaning, but there are also morphemes that have no clear meaning. In the

word cranberry the part berry is identiWable, and this makes us interpret

the word cranberry as denoting a particular kind of berry. Yet, cran- has no

particular meaning. Similarly, the Dutch compound stiefvader ‘‘stepfather’’

denotes a particular kind of father, and hence can be parsed into stief and

vader. However, the morpheme stief does not occur as a word. This

phenomenon of cranberry morphemes is widespread, and is to be expected

since complex words can lexicalize and thus survive, even though one of their

constituent morphemes has disappeared from the lexicon. The following

examples from Dutch illustrate the same phenomenon for derived words

with suYxes that are still used for coining new words (the constituent

before the suYx does not occur as a lexeme):

(7) arge-loos ‘‘naive’’, beslommer-ing ‘‘chore’’, dier-baar ‘‘dear, precious’’, le-lijk

‘‘ugly’’, moei-zaam ‘‘diYcult’’, sprook-je ‘‘fairy tale’’, veil-ig ‘‘safe’’


These recognizable suYxes determine the syntactic category of the word of

which they form a constituent. For example, -baar is a suYx that creates

adjectives, and hence dierbaar is predictably an adjective. This implies that

when we have to decompose words into morphemes, not all morphemes

have an identiWable lexical or grammatical meaning. Cranberry morphemes

like English cran- and Dutch dier- thus form a problem for an exclusively

meaning-based deWnition of the notion morpheme. This also applies to

another kind of non-aYxal bound root, the recurrent constituents of

words borrowed from Latin such as the following English verbs:

(8) conceive, deceive, perceive, receive

adduce, deduce, induce, produce, reduce

admit, permit, remit, transmit

It makes sense to consider these words complex, because of recurrent

elements such as ad-, con-, de-, in-, per-, pro-, re-, and trans- which are

preWxes, and bound roots like -ceive, -duce, and -mit. Although these bound

roots have no identiWable meaning, they should be recognized as mor-

phemes since they determine the form of corresponding noun: all verbs

in -ceive have a corresponding noun in -ception, those ending in -duce one in

-duction, and verbs in -mit one in -mission. There is a wealth of such bound

morphemes in the non-native part of the English lexicon, as the following

examples illustrate:

(9) arct-ic, cred-ible, in-del-ible, gradu-al, mor-al, mus-ic, negoti-ate, per-for-ate,

per-nic-ious

In lexeme-based morphology these bound roots do not have a lexical entry

of their own, they only occur as part of established (listed) complex lex-

emes. In morpheme-based morphology, on the other hand, they will have

to be represented as bound lexical morphemes with their own lexical entry.

The advantage of the lexeme-based approach is that it correctly predicts

that new combinations of a preWx and a bound root such as demit or

perduce are not to be expected, because we cannot assign a meaning to

such new combinations.

Boundness of morphemes is also created through allomorphy.

Allomorphy is the phenomenon that a morpheme may have more than

one shape, corresponds with more than one morph. Amorph is a particular

phonological form of a morpheme. Allomorphy is found in both aYxes

and root morphemes. In the Italian examples in (4) we saw the preWxes


al- and ap-. In fact, these are two allomorphs of the preWx ad-, in which the

Wnal consonant /d/ has assimilated to the Wrst consonant of the root

morpheme. This kind of allomorphy can be accounted for by assuming

one common underlying form /ad/ for the diVerent allomorphs of this preWx,

and a rule of assimilation that derives its diVerent surface forms.

Allomorphy is also found in root morphemes. In languages such as

Dutch, German, and Polish, obstruents (that is, stops and fricative con-

sonants, which are articulated with a high degree of obstruction in the

mouth) are voiceless at the end of a word. Hence we get alternations of

the following kind in pairs of singular and plural nouns:

(10) Dutch hoed [hut] ‘‘hat’’ hoed-en [hud

e

n]

German Tag [ta:k] ‘‘day’’ Tag-e [ta:g

e

]

Polish chleb [xlEp] ‘‘bread’’ chleb-y [xlEbØ]

The symbols between brackets represent the phonetic forms of these word

forms; the phonetic symbols are taken from the International Phonetic

Alphabet (IPA). This is the alphabet used in dictionaries and grammars to

indicate the phonetic forms of words in an unambiguous way. This is

necessary because orthographical conventions diVer from language to

language. For instance, the vowel [u] is represented as u in German, but

as oe in Dutch, as illustrated by the Wrst example in (10).

Some linguists prefer to restrict the term ‘allomorphy’ to those cases in

which the variation in phonetic shape of a morpheme does not follow from

the automatic phonological rules of the language. The alternation between

voiced and voiceless stops exempliWed in (10) is determined by a phono-

logical constraint that excludes voiced obstruents in syllable-Wnal (Dutch

and German) or word-Wnal (Polish) position. Hence, the variation in

shape of these morphemes is an automatic eVect of the phonology of the

language. This is usually accounted for by assuming a common underlying

form for the diVerent realizations of the morpheme involved, with a

morpheme-Wnal voiced obstruent. In the singular forms that lack an

overt ending, a process of syllable-Wnal or word-Wnal devoicing then

applies. The plural forms will not undergo this process because in these

forms the relevant segments do not occur in Wnal position.

This type of alternation can be contrasted to the alternation between

voiceless and voiced obstruents in English, as in the singular–plural word

pair wife–wives. This alternation applies to a small and closed set of English

words only. That is, there are alternations that are restricted to a speciWc set


of words. Another example is that the Dutch diminutive suYx has Wve

diVerent shapes (-tje, -je, -etje, -pje, and -kje); the choice of one of these

depends on the phonological composition of the stem. For instance, the

allomorph -je has to be selected after stems ending in an obstruent. The

alternations involved are unique to diminutive words, and do not follow

from general phonological constraints of Dutch. Therefore, a distinction is

made between phonology proper (the variation of the kind mentioned in

(10) that is the eVect of automatic phonological rules) and morphophonol-

ogy, the domain of phonology in which alternations are restricted to a

speciWc subset of words. The term ‘allomorphy’ might therefore be reserved

for such non-automatic alternations, which can be accounted for in

two ways. One option is to assume a common underlying form for the

allomorphs, and derive the surface forms by means of one or more

morphophonological rules, that is, rules whose application depends on

non-phonological properties such as the feature diminutive. Alternatively,

the allomorphs can be listed individually in their surface form, with

a speciWcation of the phonological context in which they occur.

In some cases the non-automatic alternation is unique for one or a few

words. For instance, the English adjective platonic, related to the noun Plato,

has the morphological structure platon-ic, with the root platon- and the suYx

-ic. The morpheme platon-, an allomorph of Plato, is a bound morpheme

since it does not occur as a word of its own. This kind of allomorphy,

a heritage from Greek (in the case of Plato) and Latin, increases the

set of bound non-aYxal morphemes enormously. An example from the

Latinate substratum of English is act, act-or vs ag-ent with the bound root

ag-. Although it has to be listed, the allomorph platon- does not require its

own entry in the lexicon: it can be speciWed in the lexical entry for Plato as

the allomorph to be used for the derivation of words from Plato by means of

non-native suYxes. The same applies to the bound root ag-.

Another, more radical form of formal variation in paradigms is the

phenomenon of suppletion, where there is no phonological similarity

between the diVerent forms of a lexeme. In the English word pair good–

betterwe observe the suppletive root bet for good, followed by the compara-

tive suYx -er. Thus, we might say that the lexeme good comprises two

diVerent stems, good and bet. In the pair bad–worse the suppletive simplex

form worse even expresses both the meaning of the stem bad and the

comparative meaning. Some linguists also use the notion ‘suppletion’ in

the domain of word-formation. In the following examples of inhabitative


names in Italian you can observe a formally regular case of derivation, a

case of allomorphy, and a case of suppletion respectively:

(11) Milano–Milan-ese, Forlı–Forliv-ese, Chieti–Teat-ino

Although alternations in the phonological shape of a morpheme may not

be the eVect of the phonology of a language, the choice of a particular

allomorph or suppletive root can still be phonologically conditioned (Car-

stairs 1988; Kiparsky 1994). For instance, the Dutch agentive suYx -aar is

selected after stems ending in the vowel [

e

] þ l, r, n (that is, in a phonolo-

gically deWned environment), and the allomorph -er elsewhere. The Italian

verb andare ‘‘go’’ has two suppletive roots: and- when the root is not

stressed, and vad- when the root is stressed in the verbal paradigm; see

(12). This example illustrates that the choice between suppletive roots may

be phonologically governed as well.


1.pers vado andiamo

2.pers vai andate

3.pers va vanno

2.2 Morphological operations

Morphology does not only deal with the analysis of existing words into their

constituent pieces. The language user is able to make new words or forms of

words, and it is this form of creativity that is the focus of morphology. The

key notion involved is that of ‘morphological operation’. This term denotes

a particular kind of linguistic activity, and invokes a dynamic perspective on

morphology. Two types of morphological operations have been discussed

so far: compounding and aYxation. They are the prototypical cases of

concatenative morphology, in which morphological constituents are concat-

enated in a linear fashion. Compounding and aYxation are the most

widespread types of morphology since they create words with a high degree

of transparency, that is, words of which the formal morphological structure

correlates systematically with their semantic interpretation.

The formal operations available in morphology have several functions.

AYxation is used both in word-formation and in inXection, and this applies

toanumberofothermorphologicaloperationsdiscussed inthis sectionaswell.


For each morphological operation, we have to deWne the set of base

words to which it applies. Often, the operation is restricted to base words of

a particular syntactic category. This is the input category of the operation.

The outputs of an operation also belong to a speciWc syntactic category.

The input category of the English suYx -able is V, and the output category

is A. Hence, verbs are the base words of the suYx -able. Thus, in the case of

derivation, the morphological operation may result in words of another

syntactic category or subcategory than that of the input words. In that case,

we speak of a category-changing or class-changing operation.

If compounding and aYxation were the only kinds of morphological

operation, morphology could be said to consist of just one operation—

concatenation. In such a view, the elements to be concatenated are lexemes

and aYxes. AYxes are provided with a subcategorization feature that

speciWes with which kind of morphological elements it has to combine.

For instance, the suYx -able will be speciWed as [V—]A, which means that it

takes verbs to form adjectives.

The reason why the term ‘morphological operation’ is more adequate

than the term ‘concatenation’ is that there are also morphological processes

that do not consist exclusively of the attachment of aYxes to words. In this

section I present a short survey of these operations, which are dealt with in

more detail in subsequent chapters on derivation and inXection.

A special kind of aYxation is the attachment of a complete or partial

copy of the base as a preWx or a suYx. This is called reduplication, illus-

trated by the following examples (Uhlenbeck 1978: 90) from Javanese:

(13) a. full reduplication:

baita ‘‘ship’’ baita-baita ‘‘various ships’’

s

e

supe ‘‘ring’’ s

e

supe-s

e

supe ‘‘various rings’’

omaha ‘‘house’’ omaha-omaha ‘‘various houses’’

b. partial reduplication:

g

e

ni ‘‘Wre’’ g

e

g

e

ni ‘‘to warm oneself by the Wre’’

jawah ‘‘rain’’ j

e

jawah ‘‘to play in the rain’’

tamu ‘‘guest’’ t

e

tamu ‘‘to visit’’

In the examples of partial reduplication, the preWx consists of a copy of the

Wrst consonant of the base followed by the vowel schwa [

e

]. The doubling

eVect of full reduplication is often reXected by its meaning contribution: for

nouns it may express plurality or distributivity (as in 13a), for verbs a high

intensity of the action expressed, and for adjectives a higher degree of the

property mentioned by the adjective.


Reduplication is a kind of aYxation (or compounding, in the case of

full reduplication), and hence to a certain extent a case of concatenative

morphology. Yet, it is clear that we cannot list reduplicative aYxes with

their phonological content in the lexicon since this content depends

on the phonological composition of the stem. The obvious analysis

is the assumption of an abstract aYx red(uplication) that triggers a

phonological operation of copying. The copy is then attached to the

copied stem.

A second type of morphological operation is the use of tone patterns.

Tone patterns belong to the suprasegmental properties of languages. In

Ngiti, the plural form of kinship terms is expressed systematically by the

tone pattern Mid–High on the stem, whatever the tone pattern of the

singular (Kutsch Lojenga 1994: 135):


aba-du aba-du ‘‘my father(s)’’

adha-du adha-du ‘‘my co-wife(s)’’

anda-du anda-du ‘‘my uncle(s)’’

Thus, we may speak of a tonal morphemeMid–High which is superimposed

on the segmental material of the stem of these nouns. This is why such a

tonal morpheme is sometimes called a supraWx. This is a case of non-

concatenative morphology since this kind of aYx is not linearly ordered

with respect to its base.

Many languages make use of internal modiWcation. Standard examples

are the patterns of vowel alternation in the roots of the so-called strong

verbs in Germanic languages, called ablaut, vowel gradation, or apophony.

Such vowel alternations are used in a number of Indo-European languages

for diVerent forms of the verb:

(15) Classical Greek: leip-o ‘‘I leave’’; le-loip-a ‘‘I have left’’, e-lipon ‘‘I left’’

The e in the Wrst root form alternates with o in the second, and zero in the

third (the second form also exhibits partial reduplication). This pattern of

vowel alternation is reXected in Germanic languages, as the following

examples from Dutch illustrate:

(16) geef [ge:f] ‘‘to give’’ gaf [gAf] ‘‘gave’’ gegeven [g ege:v e

n] ‘‘given’’

help [hElp] ‘‘to help’’ hielp [hilp] ‘‘helped’’ geholpen [g e

hOlp e

n] ‘‘helped’’

schiet [sxit] ‘‘to shoot’’ schoot [sxo:t] ‘‘shot’’ geschoten [g e

sxo:t

e

n] ‘‘shot’’


Vowel alternations also play a role in the derivation of deverbal nouns of

such verbs, as shown by the related Dutch deverbal nouns hulp ‘‘help’’ and

schot ‘‘shot’’. They only diVer from their verbal bases help and schiet with

respect to the root vowel.

Ablaut is not the only kind of vowel alternation with a morphological

function. German exhibits an alternation between back vowels and front

vowels in singular–plural noun pairs:

(17) Apfel [Apf e

l] Apfel [Epf

e

l] ‘‘apple(s)’’

Bach [bAx] Bache [bEc

e

] ‘‘brook(s)’’

Buch [bu:x] Bucher [by:c

e

r] ‘‘book(s)’’

This kind of alternation is called umlaut (also called vowel mutation or

metaphony). Historically it is a case of assimilation: back vowels of roots

are fronted before a high front vowel in the following syllable (the plural

suYx contained a high vowel originally).

If we only take the Wrst example of (17) into consideration, we might

conclude that plural formation in German is a case of non-concatenative

morphology: the plural is created by the replacement of the back root

vowel by its front counterpart. However, an alternative analysis in terms

of aYxation is also possible. Given the three examples in (17), we

might conclude that there are at least three diVerent plural suYxes in

German: ø (zero), -e, and -er. In addition, the plural nouns may exhibit

stem allomorphy, a vowel alternation triggered by the attachment of the

plural suYx. Such morphologically conditioned alternations may also aVect

consonants (Lieber 1987, 2000). English has cases of consonant modiWcation

as well, for instance defend–defence, oVend–oVence, belief–believe, and

proof–prove.

An interesting kind of non-concatenative morphology is found in,

among others, Semitic languages: root-and-pattern morphology. The basis

of each lexeme is a skeleton of consonants, in most cases three, which

functions as the root of the lexeme. The abstract pattern of consonants is

combined with one or more vowels which are intertwined with the sequence

of consonants. In addition, the lexeme may contain a preWx and a suYx. In

the words of Modern Hebrew in (18) (Clark and Berman 1984: 545) the

roots g-d-l ‘‘grow’’ and k-t-b ‘‘write’’ have been used (the k and b may

surface as x [x] and v respectively). The vowel patterns that are intercalated

with the consonantal skeletons are called transWxes since they are spread

across the consonantal sequence.


(18) Pattern Root g-d-l Root k-t-b

CaCaC gadal ‘‘grow, get bigger’’ katav ‘‘write’’

hiCCiC higdil ‘‘enlarge’’ hixtiv ‘‘dictate’’

CCiCa gdila ‘‘growth’’ ktiva ‘‘writing’’

miCCaC migdal ‘‘tower’’ mixtav ‘‘letter, missive’’

haCCaCa hagdala ‘‘enlargement’’ haxtava ‘‘dictation’’

The morphological structure of the words in (18) can be represented as

the linking between three diVerent morphemes. Each of these morphemes

forms a phonological tier of its own: (i) the skeletal tier that consists of a

pattern of consonantal and vocalic slots that is characteristic of a particular

morphological category, (ii) the sequence of consonants that represents the

lexeme, and (iii) the vowels that Wll the vocalic slots of the skeletal tier.

The words gadal and gdila in (18) can be represented as in Figure 2.1. The

consonants of the lexical root, and the vowel pattern (a-a for the base verb

and i-a for the nominalization) are both linked to the central skeletal CV

tier. These three tiers are then conXated into one sequence of sounds at the

phonetic level of the grammar, where the phonetic forms of words are

speciWed.

Fig. 2.1 Three-tiered representations of words

The morphological operations discussed so far all have the eVect that the

phonological form of the input word is changed somehow. Conversion, on

the other hand, consists of a change in syntactic (sub)category only. The

conversion of nouns to verbs is quite common in European languages; see

(19). The verbs are given here in their citation form, the inWnitive. The

conversion from noun to verb is not indicated directly by means of an aYx,

and is therefore also called implicit transposition, as opposed to explicit

transposition, which denotes cases of category-changing word-formation in

which the change is marked through the addition of an aYx. Note that

conversion does have indirect morphological eVects: the verbs in (19) are

recognized as such by their verbal inXectional endings, the inWnitival


suYxes (except the English verb since there is no overt inWnitival ending in

English). The category change may also have an eVect on the stress pattern,

as in the English pair convert (V)—convert (N), where the noun is derived

from the verb, with concomitant stress shift from the last to the Wrst syllable.

(19) Noun Verb

Dutch Wets ‘‘cycle’’ Wets-en ‘‘cycle’’

English chain (to) chain

French guide ‘‘guide’’ guid-er ‘‘guide’’

Latin corona ‘‘crown’’ coron-a-re ‘‘crown’’

If one wants to treat conversion as a kind of aYxation, one is forced to

assume a zero-morpheme that is added to the input word. However, there is

no independent evidence for such a zero-aYx, and we do not even know if

the zero-morpheme should be taken to be a preWx or a suYx. Therefore,

conversion as exempliWed in (19) is better analysed in terms of the following

morphological rule:

(20) [x]N ! [[x]N]V

The verbs in (19) therefore have the structure [ [X]N]V.

A deWning property of the notion ‘conversion’ is that it has a direction:

in the examples above, the verb has been derived from the noun. This

phenomenon must therefore be distinguished from multifunctionality, the

situation inwhichwords can be used for diVerent syntactic categories without

a particular direction in the relation between these diVerent uses of words. In

Maori, for example, thewordwaiata canbeusedas a verb ‘‘to sing’’, as a noun

‘‘song’’, and as a participle ‘‘singing’’ (Bauer 1993: 510). In Sranan, a creole

language of Surinam, theword hebi functions as an adjective ‘‘heavy’’, a noun

‘‘weight’’, an intransitive verb ‘‘to be heavy’’, and a transitive verb ‘‘to make

heavy’’ (Voorhoeve 1979: 43).

Change of category without overt morphological marking is also found in

the case ofmiddle verbs, which are intransitive anddenote a property, whereas

the corresponding activity verb denotes an activity (examples from Dutch):

(21) Deze aardappelen schillen gemakkelijk

These potatoes peel easily

‘‘These potatoes are easy to peel’’

Mars hapt zo heerlijk weg


Mars eats so nicely away

‘‘Mars is so pleasant to eat’’

(Mars is a kind of candy bar.) This kind of change from one subcategory of

verbs to another subcategory may be subsumed under conversion because

there is a clear direction in the relation between the verbs involved: the

middle verb is derived from the activity verb.

In Chapter 1 you were introduced to the notion of paradigmatic word-

formation, in which a morphological constituent of a word is replaced with

another one. A typical case of this kind of word-formation is aYx substi-

tution, the replacement of one aYx with another. In Dutch, female

counterparts of agent nouns can be formed by replacing -er with -ster

(Booij 2002a: 6):

(22) aanvoerd-er ‘‘captain’’ aanvoerd-ster

betwet-er ‘‘lit. better knower, pedant’’ betweet-ster

rederijk-er ‘‘rhetorician’’ rederijk-ster

reizig-er ‘‘traveller’’ reizig-ster

oproerkraai-er ‘‘ring leader’’ oproerkraai-ster

The operation of substitution as a viable way of making new words has

developed from systematic relationships between words derived from the

same base. In this case, both -er and -ster can be added to Dutch verbs to

form agent nouns. Thus a pattern [X-er]N: [X-ster]N could be observed,

which was then extended to other nouns in -er without a straighforward

verbal base. For instance, there is no Dutch verb reizig ‘‘to travel’’, and yet,

the agent noun reiziger has a female counterpart in -ster. The presence of

the /d/ in aanvoerdster also betrays that this word is derived from aanvoer-

der. The /d/ does not belong to the verbal stem aanvoer ‘‘to lead’’, but is part

of the allomorph -der that is used after stems ending in /r/. Since it is -er that

is replaced, the /d/ shows up in the female agent noun as well.

A prototypical case of paradigmatic word-formation is back formation in

which the direction of derivation is inverted: the less complex word is

derived from the more complex word by omitting something. Well-known

examples from English are to sculpt from sculptor, and to babysit from

babysitter. The noun sculptor is a borrowing from Latin. Because English

has word pairs of the type V�[Vþor]N, (terminate–terminator, etc.),

the verb sculpt could be reconstructed from the noun sculptor by reinter-

preting this word as having the structure [ [sculpt]Vor]N. The paradigmatic

dimension involved here is that a word ending in -or is assigned an internal


morphological structure with a verbal base on the basis of existing

verb–noun pairs such as terminate–terminator.

The emergence of the verb to babysit can be reconstructed as follows.

The word babysitter is a regular compound consisting of two nouns, baby

and the deverbal noun sitter. However, there is no general process of NþV

compounding in English. The exceptional NV compound babysit could

therefore only arise through back formation. In the same way, theDutchNV

compound stofzuig ‘‘to vacuum-clean’’ arose through back formation from

the regular NN compound stofzuiger ‘‘lit. dust sucker, vacuum cleaner’’.

In the cases of to babysit and stofzuig, the structure [ [N] [V -er]N]N has

been reinterpreted as [ [N V]V -er]N, and subsequent back formation led

to the rise of these N þ V compounds.

2.3 Morphological typology

The catalogue of morphological operations presented in section 2.2 raises

the question to what extent the languages of the world make use of these

possibilities. First, we can locate each language on a scale of degree of

synthesis, the average number of morphemes in a word. On one end of the

scale we Wnd isolating languages that do not make use of morphology at all.

A classical example of such a language is Vietnamese (which, however, is

said to have compounds). At the other end of the scale we Wnd polysynthetic

languages such as Greenlandic and Alaskan Yup’ik, languages in which

words may contain a considerable number of suYxes after the root.

Before we have a look at some relevant examples, I will Wrst give a short

clariWcation of the notational conventions used in interlinear morphemic

translation. These conventions are of considerable importance for our

understanding of the structure of sentences and words. A space marks the

boundary between twowords, and a hyphen represents a boundary between

two morphemes within one word. Lexical morphemes are represented by

lower case letters, and grammatical categories by small capitals. If one

morpheme on the Wrst line represents more than one piece of lexical or

grammatical information in the morphemic gloss, the categories are separ-

ated by a dot, as in the following Latin example (Lehmann 1982: 205):

(23) Manu-s manu-m lava-t

hand-nom.sg hand-acc.sg wash-3sg

‘‘One hand washes the other’’


The only exception to this use of the dot is its absence in combinations of

the category person and the category number, as in 3sg.

Let us now have a look at some examples of polysynthetic words in

which the conventions just discussed are also exempliWed—Wrst Green-

landic (Fortescue 1984: 273) and then Alaska Yup’ik (Mithun 1999: 28):

(24) tuqu-riikatap-puq

die-long.ago–3sg.indic

‘‘He died long ago’’

anglani-tu-llru-u-nga caknek

enjoy-customarily-past-ind.intr-isg very.much

‘‘I used to enjoy myself very much’’

The Wrst example is a sentence of one word only, the second one contains

two words. When we compare this with the number of words in the English

glosses (four and seven respectively), we get some idea of what it means for

a language to be polysynthetic.

The second scale on which we may rank languages as to their morpho-

logical properties is that of fusion. In some languages, a word is easily

segmentable into its constituent morphemes. An example is Turkish, a

language that is therefore characterized as agglutinative: the stem of

a word is followed by one or more suYxes, each with their own meaning:

(25) cocuk-lar-nız-dan

child-pl-your.pl-abl ‘‘from your children’’

Most Indo-European languages are fusional in their inXectional system

since diVerent inXectional properties are often expressed by one and the

same morpheme. In the Polish word form koty ‘‘cat, nom.pl’’ the ending -y

expresses simultaneously the properties nominative and plural. In the

English word walks, the ending -s expresses three properties: present

tense, singular, and 3.person. Such units that serve to express more

than one morphological property are called portmanteau morphs.

When a language tends to be more agglutinative, it will tend to have

more morphemes per word than a fusional language, and hence it will be

higher on the scale of synthesis as well. The average number of morphemes

per word in Turkish is estimated to be four times higher than that in

English (Csato and Johanson 1998: 208).

The kind of typological classiWcation discussed so far mainly has a

descriptive and orientational function: by locating a language on a number


of scales, we know roughly what kind of morphological system we may

expect. But it does not provide Wne-grained classiWcations. Germanic lan-

guages are fusional in their inXectional systems, but agglutinative in their

system of derivational word-formation. Moreover, for an adequate de-

scriptive classiWcation other parameters are also relevant, for instance the

parameter of reduplication: languages of the Austronesian family make

wide use of reduplication patterns, whereas this does not apply to most

Indo-European languages of Europe. Languages may also diVer in the

extent to which they make use of preWxation or suYxation.

Morphological typology becomes theoretically interesting if it enables us

to predict certain properties of a language on the basis of other properties.

For instance, the following morphological universal has been proposed by

Greenberg (1963: 95):

(26) If a language has the category of Gender, it always has the category of

Number.

This universal has the form of an implication, and hence it predicts that of

the four following logically possible languages only the Wrst three exist:

(27) a. Languages with Gender and Number

b. Languages with Number only

c. Languages without Gender or Number

d. Languages with Gender only

Thus, an implicational universal is a restriction on the class of possible

natural languages, and hence contributes to the deWnition of the notion

‘possible natural language’.

Some implicational universals pertain to markedness phenomena.

Markedness is the asymmetrical distribution of properties. An example

of a markedness universal is that there are many languages in which the

singular is not expressed by a morpheme, but only the plural, whereas there

are no languages where only the singular is expressed by a morpheme. For

example, the asymmetric distributional pattern of singular and plural

morphemes given in Table 2.1 has been found. This table shows that

languages with a singular morpheme only must be excluded in order

to restrict the degree of variation in natural language. Hence, we might

formulate the following implication universal for this markedness pattern:

‘if singular number is expressed by a morpheme, then plural number

as well’.


This generalization concerning the expression of number has to be

amended slightly, however. There are languages where, for those entities

that always occur in pairs or in groups, the plural form of the noun has no

overt suYx, and the singular form ends in a singulative suYx. This is

the case for Turkana (Dimmendaal 1983: 224–8), a language of Kenya

with the singulative suYxes -a and -it (the preWxes are gender markers):

(29) singulative plural

E-sØkØn-a ‘‘breast’’ N-sØkØn ‘‘breasts’’

e-turkana-Øt ‘‘Turkana person’’ N-turkana ‘‘Turkana people’’

This reversal of the markedness pattern concerning singular–plural in a

special domain is called local markedness (Tiersma 1982).

The use of hierarchies in morphological typology is illustrated by the

following hierarchy for the diVerent values for the category number:

(30) singular > plural > dual

This hierarchy ranks singular above plural, and plural above dual. It

expresses that singular forms are less marked than plurals, and plurals

are less marked than duals. This means that if a language has a dual (that

is, a word form with 2 as the value for the category number), it has also a

plural, and if a language has a plural, it also has a singular. Hence, this

hierarchy restricts the variation space of natural language: certain types of

logically possible languages are excluded, such as a language with singular

and dual only.

Not all the typological universals are absolute ones; some are statistical

tendencies only. For instance, there are many more languages that only

use suYxes (Turkish is an example) than there are languages that only use

preWxes. Hence, there is a suYxing preference in natural languages. Yet,

there are languages that are exclusively preWxing, so there is no absolute

universal involved here. Many of the universals discussed in Greenberg

(1963) are of this statistical nature.

Table 2.1. Distribution of Number morphemes

Absence of sg morpheme Presence of sg morpheme

Presence of pl morpheme English, Dutch Latvian, Italian

Absence of pl morpheme Chinese, Maori

Source: Croft 1990: 69.


Summary

Words can be divided into diVerent kinds of morphemes such as roots and

aYxes, the morphological atoms of language. These morphemes may vary

in shape (allomorphy), a variation that does not always follow from the

phonological system of the language. In the case of suppletion, diVerent

stems co-occur in the paradigm of one lexeme.

The set of morphological operations available to human languages

comprises more than concatenation: conversion, reduplication (concaten-

ation plus copying), diVerent types of phonetic modiWcation, root-and-

pattern morphology, and paradigmatic word-formation also play a role.

Languages do not all make the same use of the available morphological

operations, and can be classiWed according to the indices of synthesis and

fusion. In addition to this purely classiWcatory typology, morphologists

make cross-linguistic comparisons of morphological systems in order to Wnd

constraints on the degree of morphological variation of natural language.

Questions

1. Identify the bound constituents of the following English words: disagree-able, acceptability, ungrammaticality, discriminatory, permafrost, fascination,protolanguage, versification, intolerance, productivity, unidirectionality.

2. Consider the sets of morphologically related words in French in the tablethat exhibit variation in the underlined vowel of their base word (Dell andSelkirk 1978).

Base word Derived words

[œ] [œ] [ c]fleur ‘‘flower’’ fleurette ‘‘small flower’’ floral ‘‘floral’’seul ‘‘alone’’ seulement ‘‘only’’ solitude ‘‘solitude’’peuple ‘‘people’’ peuplade ‘‘tribe’’ populaire ‘‘popular’’[E] [E ] [a]vain ‘‘idle’’ vainement ‘‘in vain’’ vanite ‘‘vanity’’clair ‘‘clear’’ eclairer ‘‘to light’’ clarifier ‘‘to clarify’’mer ‘‘sea’’ amerrir ‘‘to land on the sea’’ marin ‘‘sailor’’africain ‘‘African’’ Africaniste ‘‘Africanist’’humain ‘‘human’’ humanite ‘‘humanity’’similaire ‘‘similar’’ similarite ‘‘similarity’’


a. Formulate the rule that accounts for these vowel alternations.b. Is this rule an automatic phonological rule or a morphophonological

rule?

3. In the following past tense forms of English verbs, the past tense suffix hasthree different phonetic shapes: kept, walked, kissed, hugged, lived, added,fitted, coded.

a. Which are the three phonetic forms of this suffix?b. Formulate the rules (or rule) that account for this phonetic variation.

4. Make a morphological analysis of the following words of the Amerindianlanguage Cree, and give the interlinear morphemic glossing of the lastword (Cowan and Rakusan 1985: 111):

niwa:pahte:n ‘‘I see (it)’’kiwa:pahte:n ‘‘you see (it)’’niwa:pahte:na:n ‘‘we see (it)’’kiwa:pahte:na:wa:w ‘‘you (plural) see (it)’’nima:cise:n ‘‘I cut (it)’’kima:cise:n ‘‘you cut (it)’’nima:cise:na:n ‘‘we cut (it)’’kima:cise:wa:w ‘‘you (plural) cut (it)’’nitapin ‘‘I sit’’kitapin ‘‘you sit’’nitapina:n ‘‘we sit’’kitapina:wa:w ‘‘you (plural) sit’’

5. Consider the following past tense forms of English: kept, wept, slept. Whichkinds of operation have been used for making these verbal forms?

6. In written Arabic, verbs can be derived from nouns, as the following ex-amples (Becker 1990b: 14) show:

zayt ‘‘oil’’ zayyata ‘‘to oil’’bukla ‘‘buckle’’ bakkala ‘‘to buckle’’turki: ‘‘Turk’’ tarraka ‘‘to Turkify’’tilifu:n ‘‘telephone’’ talfana ‘‘to telephone’’tilifiziu:n ‘‘television’’ talfaza ‘‘to televise’’

a. Give a three-tiered representation of the verbs bakkala and talfana.b. Give the consonantal template and the vowel template for the mor-

phological category of denominal verbs.

7. The following are singular and plural nouns in Bulgarian (Cowan andRakusan 1985: 99):

teat er teatri ‘‘theater(s)’’bob er bobri ‘‘beaver(s)’’pesen pesni ‘‘song(s)’’psalom psalmı ‘‘psalm(s)’’bancik bancigi ‘‘band saw(s)’’ızverk ızvergi ‘‘monster(s)’’


Give the (morphological and phonological) rules for plural noun formationfor this set of word forms.

8. Which types of morphological processes are involved in the followingforms (Melcuk 2000: 528) of the Tagalog verbs patay ‘‘to kill’’ and sulat‘‘to write’’?

9. Consider the following pairs of singular and plural nouns of Agta (Wiltshireand Marantz 2000: 558):

takki taktakki ‘‘leg(s)’’labang lablabang ‘‘patch(es)’’uffu ufuffu ‘‘thigh(s)’’

Formulate the morphological rule that accounts for the formation of theplural nouns.

10. In Italian, adjectives can be derived from nouns and adjectives through theaddition of the suffix -oso or -astro, as illustrated by the following examples(source Scalise 1984: 59):

fama ‘‘fame’’ famoso ‘‘famous’’virtu ‘‘virtue’’ virtuoso ‘‘virtuous’’giallo ‘‘yellow’’ giallastro ‘‘yellowish’’blu ‘‘blue’’ bluastro ‘‘bluish’’

Give the stem-forms of the four base words listed here.

Further reading

A survey of the different types of morphological operations can be foundin chapter 8 of BLM, and in Payne (1997). A detailed analysis of vowel andconsonant mutation is given in Lieber (1987). The multi-tier interpretationof root-and-pattern morphology has been proposed by McCarthy (1981). Thiskind of morphology is also found in a number of non-Semitic languages(Broselow 2000).Formal theories of morphology as a set of operations on lexemes (processual

morphology) have been developed in Anderson (1992) and Stump (2001).Morphological typology is discussed in Comrie (1981) and Croft (1990).

The suffixing preference is dealt with in Cutler et al. (1985). The verbs of theAthapaskan language Slave are famous for their morphological complexity

past present

active pumatay pumapataysumulat sumusulat

passive pinatay pinapataysinulat sinusulat


(Rice 1989, 2000), and Amerindian languages in general for their polysyn-thetic nature (Mithun 1999). Polysynthesis is discussed from a theoretical pointof view in Baker (1988, 1996, 2001). Possible explanations for the suffixingpreference are discussed in Hawkins and Cutler (1988), and in Hall (1988).Information on the World Atlas of Language Structures that also includes

morphological typology can be found at http://linguistics.buffalo.edu/people/faculty/dryer/dryer/atlas. The Universals Archive of the University ofKonstanz may be consulted at http://ling.uni-konstanz.de/pages/proj/sprachbau.htm. The website of the Surrey Morphology Group provides typo-logical information on agreement and syncretism: http://www.surrey. ac.uk/LIS/SMG.A standard for interlinear morphemic glossing is presented in Lehmann

(1982). More recent versions are the set of glossing rules proposed by theLeipzig typologist group, based on Lehmann (1982), which can be found atwww.eva.mpg.de/lingua/index.html, and Lehman (2004).


http://linguistics.buffalo.edu/people/faculty/dryer/dryer/atlas

http://linguistics.buffalo.edu/people/faculty/dryer/dryer/atlas

http://ling.uni-konstanz.de/pages/proj/sprachbau.htm

http://ling.uni-konstanz.de/pages/proj/sprachbau.htm

http://www.surrey.ac.uk/LIS/SMG

http://www.surrey.ac.uk/LIS/SMG

www.eva.mpg.de/lingua/index.html

Part IIWord-Formation


3Derivation

3.1 Lexeme formation

The basic function of derivational processes is to enable the language user

to make new lexemes. Lexemes belong to lexical categories such as N, V,

and A and the derived lexemes may belong to a diVerent category than their

bases. The examples in (1) from Dutch illustrate the possible categorial

shifts, and also cases in which the lexical category does not change (Booij

2002a: 87). Words are divided into two kinds of lexical classes: open and

closed classes. In most languages, nouns, adjectives, and verbs form open

classes. As illustrated in (1), these classes can be extended by means

of word-formation. Function words such as determiners, conjunctions,

pronouns, and adpositions (pre- and postpositions) form closed sets

of words that cannot be extended by regular word-formation patterns.

The base words that form inputs to word-formation are normally also

words of these open classes, but there are exceptions. For instance, the

Dutch diminutive suYx -je can be attached to the demonstratives dit ‘‘this’’

and dat ‘‘that’’, as in dit-je-s en dat-je-s ‘‘odds and ends’’, and to phrases

such as the PP onder ons ‘‘between us’’, with the corresponding diminutive

onderons-je ‘‘private chat’’. This extension of the input domain to function

words and phrases is typical for highly productive word-formation

processes.

3.1 Lexeme formation 51

3.2 Templates and idiosyncrasies 61

3.3 Constraints on derivation 64

3.4 Productivity 67

3.5 AYx ordering 71

Summary 72

Questions 73

Further reading 74

(1) Derivation of nouns

A ! N suYxation schoon ‘‘beautiful’’ schoon-heid ‘‘beauty’’

V ! N suYxation spreek ‘‘to speak’’ sprek-er ‘‘speaker’’

preWxation praat ‘‘to talk’’ ge-praat ‘‘talking’’

N ! N suYxation moeder ‘‘mother’’ moeder-schap ‘‘motherhood’’

preWxation zin ‘‘sense’’ on-zin ‘‘nonsense’’

Derivation of adjectives

N ! A suYxation meester ‘‘master’’ meester-lijk ‘‘masterly’’

V ! A suYxation lees ‘‘to read’’ lees-baar ‘‘readable’’

A ! A suYxation blauw ‘‘blue’’ blauw-ig ‘‘blueish’’

preWxation gewoon ‘‘common’’ on-gewoon ‘‘uncommon’’

Derivation of verbs

N ! V suYxation analyse ‘‘analysis’’ analys-eer ‘‘to analyse’’

preWxation slaaf ‘‘slave’’ ver-slaaf ‘‘to enslave’’

A ! V suYxation kalm ‘‘calm’’ kalm-eer ‘‘to calm down’’

preWxation bleek ‘‘pale’’ ver-bleek ‘‘to turn pale’’

V ! V suYxation krab ‘‘to scratch’’ krabb-el ‘‘to scratch lightly’’

preWxation rijd ‘‘to ride’’ be-rijd ‘‘to ride on’’

Many languages have a fourth open lexical class, that of adverbs. This class

can also be extended in regular ways. In English, adverbs can be derived from

adjectives by means of the suYx -ly, and in French by means of suYxation

with -ment, as in lente-ment ‘‘slowly’’. Nouns are the bases of adverbs such as

English group-wise, and Afrikaans regerings-weee ‘‘by the government’’

derived from regering ‘‘government’’. Adverbs are special in that they tend

not to feed other word-formation processes, unlike words of the three other

open categories. InXection of adverbs is also rare, but the comparative form

of an English adverb such as sooner suggests that inXection is not completely

excluded (note that soon functions as an adverb only, not as an adjective).

The description of each process of derivation consists of a speciWcation of the

properties of the input words, and those of the output words. The Wrst set of

examples in (1) consists of cases of nominalization (the derivationof nouns from

words of other word classes). Such derivation processes usually impose con-

straints on the lexical category of their input words, for instance that they have

to be adjectives. Besides nominalization, we distinguish verbalization, adjecti-

valization, and adverbialization as types of category-determining processes.

52 word-formation

All derivational suYxes in (1) appear to determine the lexical category

of the complex words that they form, and hence, they are category-

determining. This generalization holds for all Germanic languages. For

instance, it is the diminutive suYx -(t)je of Dutch that determines the lexical

category of diminutives, which are always nouns. This suYx does not only

attach to nouns, but also to adjectives and verbs. In the latter cases the

derived words are also nouns: blond ‘‘blond’’ (A)—blondje ‘‘blond girl’’

(N), speel ‘‘to play’’ (V)—speeltje ‘‘toy’’ (N).

A suYx may be category-determining even if it does not change the

lexical category of its input words. When Dutch diminutive nouns are

derived from nouns, they do not change the category. Yet, we can see

that they are category-determining because they always have neuter gender

even if the input noun is non-neuter, as illustrated by the word pair

zaal ‘‘room, non-neuter’’—zaaltje ‘‘small room, neuter’’. That is, the

diminutive suYx determines the lexical subclass of its derived words.

This category-determining role of suYxes has led some linguists to

extend the notion head, an important notion in syntax, to the domain of

morphology. In syntax, phrases consist of minimally a head, and it is the

syntactic category of the head that determines the category of the phrase.

A phrase with an adjectival head is an Adjective Phrase (AP), a phrase with

a noun as head is a Noun Phrase (NP), etc. For instance, very good is an AP

with the adjective good as its head, and hard work is an NP with the noun

work as its head. By analogy, we may say that the Dutch diminutive suYx is

the head of the diminutive noun since it determines the syntactic category

of this type of complex word. This morphological use of the notion ‘head’

has been advocated in particular in the morpheme-based approach

to morphology. In that framework, aYxes get their own lexical entry in

the lexicon, in which their class-determining properties are speciWed.

The morphological structure of the Dutch diminutive noun blondje might

then be represented as in Figure 3.1. The lexical category N and the gender

Fig. 3.1 Percolation of head features

derivation 53

feature [+neuter] are taken to be properties of the suYx -je, and percolated

to the dominating node, thus classifying this word as a neuter noun. The

percolation procedure will percolate the properties of the head to the dom-

inating node of the whole word (Lieber 1980, 1989). Additional evidence for

the headship of the Dutch diminutive suYx is that it governs the selection of

the correct plural suYx. Whereas simplex nouns ending in [

e

] take either -s

or -n as their plural suYx, the diminutive nouns in -je [

e

] always select -s.

Williams (1981: 248) proposed the following rule for the identiWcation of

the head constituent of complex words:

(2) Right-hand Head Rule (RHR): In morphology, we deWne the head of a

morphologically complex word to be the right-hand member of that word.

This rule will identify suYxes as heads, and hence, they will determine the

lexical (sub)category of the words they create. The rule is supposed to be

applicable to both derived words and compounds. In English compounds it

is indeed the lexical category of the right constituent that determines the

lexical (sub)category of the compound: a bottle factory is a kind of factory,

not a kind of bottle. Similarly, English derivational suYxes are category-

determining. However, the Right-hand Head Rule cannot be a universal

rule because there are languages with left-headed complex words, as

illustrated in section 4.1.

You should realize that the notion ‘head’ as used here is diVerent from its

standard use in syntax. In syntax, a head is not identiWed by its linear

position in the phrase, but by the hierarchical conWguration in which it

occurs. Moreover, the RHR is at best a language-speciWc principle since

there are also languages with category-determining preWxes or category-

neutral suYxes. In the list (1) given above we see examples of both

category-neutral preWxes (the negative preWx on- that derives adjectives

from adjectives, and nouns from nouns), and of category-determining pre-

Wxes such as ver-that derives verbs from adjectives. The latter preWx thus

forms a problem for assuming theRHR to be generally applicable toDutch.

There are many other languages with category-determining preWxation.

The generalization expressed by the RHR has a historical explanation.

In languages with right-headed phrases, the univerbation of such phrases

has led to the emergence of right-headed compounds. Subsequently, some

right constituents of compounds have developed into suYxes. The English

suYx -dom (as in kingdom) with the meaning ‘‘domain’’ derives from the

Old English word dom ‘‘fate, destiny’’, and the Dutch suYx -loos ‘‘-less’’

54 word-formation

(as in hulpe-loos ‘‘helpless’’) derives from the adjective loos ‘‘being with-

out’’, words that once occurred in the head position of compounds. Thus,

such suYxes could start functioning as morphological heads. On the other

hand, preWxes often derive from words functioning as left, non-head con-

stituents of compounds or from preverbal adverbs. The English preWx over-

(as in to overdo) derives historically from the word over and the Latin preWx

ab- (as in abducere ‘‘to take away’’) from the preposition ab. Due to their

origin from words in a (morphological or syntactic) non-head position,

such preWxes do not function as heads either (Hall 1991).

The category-neutral evaluative suYxes of Italian and other Romance

languages also pose a problem for the universal applicability of the RHR

(Scalise 1988). The Italian diminutive suYx -ino, for example, derives

nouns from nouns, but adjectives from adjectives, and adverbs from

adverbs. Hence it is category-neutral as illustrated by the following

examples: tavolo ‘‘table’’—tavolino ‘‘small table’’, giallo ‘‘yellow’’—giallino

‘‘yellowish’’, bene ‘‘well’’—benino ‘‘reasonably well’’. That is, these suYxes

do not function as heads.

In lexeme-based morphology, the category-determining nature of a

particular aYx can be expressed in the template for that aYx. The Dutch

diminutive suYx will thus occur in the following template:

(3) [[x]Y tje]N ‘‘small entity with Y-related property’’ where Y ¼ N, A, V

The variable Y used here stands for the lexical categories, N, A, and V.

Words from these categories may serve as base words (and a few other

words and phrases as well). The output category is always N. The Dutch

category-neutral preWx on- ‘‘un-’’ will be speciWed as follows:

(4) [on[x]Y]Y ‘not-Y’, where Y ¼ N, A

This template expresses that on- is a preWx, not a lexeme since the preWx

does not carry a category label of its own. It speciWes that the preWx on-

takes nouns and adjectives as base words, and that it is category-neutral

since the variable Y recurs as category of the whole word.

Morphological rules do not necessarily apply to existing lexemes: pos-

sible words may also function as inputs. Dutch negative adjectives preWxed

with on- ‘‘un-’’ often have a possible, but non-existing deverbal adjective

ending in -baar ‘‘-able’’ as base word. Similar observations can be made for

English adjectives in -ible and -able: undefatigable is an existing word,

unlike its base word defatigable.

derivation 55

(5) aanraakbaar ‘‘touchable’’ onaanraakbaar ‘‘untouchable’’

uitstaanbaar ‘‘tolerable’’ onuitstaanbaar ‘‘intolerable’’

verwoestbaar ‘‘destroyable’’ onverwoestbaar ‘‘undestroyable’’

corrigible incorrigible

delible indelible

defatigable indefatigable

The positive adjectives in -baar, -ible, and -able in the left column of (5) do

not occur by themselves. However, intuitions about existence versus non-

existence of a particular positive adjective may vary from speaker to

speaker, and sometimes the positive adjective crops up in usage. The crucial

point is that existence is not a necessary condition for these adjectives to be

used as base words. In the case of Dutch, we cannot assume, as a way out

for this problem, that the preWx on- has been attached before the suYx

-baar, because the preWx on- does not attach to verbs. It is the adjectival

suYx -baar that makes the attachment of this preWx possible.

We might interpret these facts, the co-occurrence of two word-formation

processes, as showing that the use of onemorphological operationmay imply

the use of a second one. A formal representation of this co-occurrence is

the assumption of a conXated morphological template [on[V-baar]A]A, a

conXation of the template [on-A]A and the template [V-baar]A. This simul-

taneous attachment of a preWx and a suYx is called parasynthetic word-

formation. The formation of Italian verbs from adjectives by adding both a

preWx and a thematic vowel to an adjectival stem might be seen as another

instance of this kind of word-formation (Scalise 1984: 146):

(6) brutt-o ‘‘ugly’’ im-brutt-i-re ‘‘to make ugly’’

rozz-o ‘‘crude’’ di-rozz-a-re ‘‘to make less crude’’

vecchi-o ‘‘old’’ in-vecchi-a-re ‘‘to age’’

Note that this kind of word-formation is diVerent from circumWxation. In

the latter case we have to do with one, discontinuous aYx, whereas in the

examples in (6) two independently occuring aYxes are used simultaneously.

Rules or templates for derivational patterns describe both the formal

(phonological and syntactic) properties and the semantic properties of the

class of output words. The semantic interpretation of a complex word is

determined by the template, and includes the meaning of the base word.

Hence, the meaning of a complex word is in principle a compositional

function of the meaning of its base word and its morphological structure,

as illustrated by the templates (3) and (4).

56 word-formation

The meaning contribution of the aYx, or more generally, the deriv-

ational process, might actually be pretty vague. English (and many other

languages) make use of a number of suYxes to derive relational adjectives

from nouns:

(7) Noun NP with relational adjective

atom atomic bomb

music musical performance

ocean oceanic winds

parent parental refusal

president presidential speech

The suYxes -ic, -al and -ial are used to formdenominal adjectives that serve to

express that there is some relation between the base noun of the adjective and

the head noun of the phrase. The precise nature of that relationship is not

speciWed, and needs further interpretation by the language user. In the case of

parental refusal, the parents are engaged in refusing, inmusical performance it

is the music that is being performed. The same semantic vagueness in

the relation between base word and derived word can be observed when

we compare the diVerences in meaning between the following denominal

verbs ending in -ize: hospitalize ‘‘to put someone in hospital’’, burglarize

‘‘to do the work of a burglar’’, vaporize ‘‘to turn something to vapour’’.

Conversion can be subsumed under derivation, although there is no

phonological change involved, because it serves to coin new lexemes on

the basis of existing ones. The examples in (8) illustrate the conversion of

nouns to verbs.

(8) Noun Verb

English bed bed ‘‘to go to bed’’

bottle bottle ‘‘to put into a bottle’’

bomb bomb ‘‘to throw bombs on’’

Dutch kaas ‘‘cheese’’ kaas ‘‘to produce cheese’’

melk ‘‘milk’’ melk ‘‘to extract milk from’’

stof ‘‘dust’’ stof ‘‘to remove dust from’’

Other input categories involved in conversion in English are adjectives and

verbs:

(9) A > V: calm > to calm, clean > to clean, wet > to wet

V > N: to cheat > (a) cheat, to take > (a) take, to approach > (an) approach

derivation 57

How can we be certain about the direction of the conversion given the fact

there are no phonological indications for this direction? Arguments

for determining the direction are often (but not always) provided by the

semantics of the conversion pair. In the examples in (8), the meaning of the

verb is a compositional function of that of the noun, and that is why

we interpret the relation in terms of verbs being derived from nouns.

The phonological make-up of words may also form an indication. Dutch

simplex verbs consist of either one syllable, or two syllables, the second of

which contains a schwa. Yet, we Wnd verbs such as papegaai ‘‘to imitate’’

and domino ‘‘to play dominoes’’ that do not exhibit this phonological make-

up since they consist of three syllables. This can be explained by these verbs

being conversions of the nouns papegaai ‘‘parrot’’ and domino ‘‘dominoes’’

respectively, which accounts for their marked phonological structure.

The template for the conversion cases in (8) will be as follows:

(10) [[X]N]V ‘‘to perform an action in which [X]N plays a role’’

Again, the semantic description is vague; the precise role of the N, and

hence the nature of the action expressed, is not deWned exactly. These are

determined by non-grammatical knowledge. Thus, the actual meaning of

the conversion verbs may be quite varied, which increases the semantic

Xexibility, and hence the usability of this pattern of lexeme formation.

The interpretation of conversion as a case of lexeme formation is sup-

ported by the observation that languages may diVer in that one language

uses an aYx to coin a verb from a noun, whereas another language creates

the corresponding lexeme by means of conversion. For instance, the Dutch

de-adjectival conversion verb wit ‘‘to make white’’ corresponds to English

whiten, coined by addition of the suYx -en to the adjectival stemwhite. Even

within a single language one may Wnd both types exempliWed; compare the

English verbs to wet and to dampen to their adjectival bases wet and damp.

The coinage of middle verbs is another instantiation of lexeme formation

without overt phonological eVect. Middle verbs express a generic property

and hence are stative verbs, whereas their source verbs indicate an action.

This process is used in Dutch and English, whereas Russian requires a suYx

-sja to perform the same semantic operation (Spencer andZaretskaya 1998):

(11) Dutch Vurenhout zaagt gemakkelijk

Pine saws easily

‘‘Pine saws easily’’

58 word-formation

Russian Sosna legko pilit-sja

Pine easily saw-SJA

‘‘Pine saws easily’’

Middle verbs often require the presence of an adverbial expression such as

easily: a sentence like Pine saws, without an adverbial expression, is odd.

This can be explained by the Non-Redundancy Constraint (Ackerman

and Goldberg 1996: 21; Goldberg and Ackerman 2001) that forbids

morphological operations expressing redundant information. Since you

know that pine is a kind of wood that can be sawn, it does not make

sense to state this, unless you add a modiWcation with some informational

content to that statement, in this example the adverb easily. For the

same reason, an adjective such as eyed does not make much sense as a

speciWcation of the properties of human beings since they normally have

eyes, whereas the adjective blue-eyed expresses a non-redundant piece of

information about human beings. This illustrates that there is a diVerence

between the well-formedness of a complex word, and its semantic or

pragmatic appropriateness.

The interpretational versatility that we observed for relational adjectives

can also be found in the case of reduplication. Reduplication is often

(but not always!) a case of iconicity in language. This means that the formal

structure is iconic for the meaning expressed. This applies particularly

to full reduplication in which the form is doubled; the corresponding

meaning also appears to be doubled in the sense that it may express

repetition, intensity, plurality, distributivity, etc. Afrikaans has borrowed

reduplication, an unusual word-formation process in Germanic languages,

from Malay, and this process is used frequently (Botha 1988: 1–2):

(12) Die kinders drink bottels-bottels limonade

the children drink bottles-bottles lemonade

‘‘The children drink bottles and bottles of lemonade’’

Die leeu loop brul-brul weg

the lion walk roar-roar away

‘‘Roaring repeatedly, the lion walks away’’

Hy dra tien-tien boek die trap op

he carry ten-ten books the stairs up

‘‘He carries the books up the stairs ten at a time’’

Derivation is not only used in a category-changing fashion, but also to

create other semantic subcategories of the same lexical category. Italian has

derivation 59

a lot of category-neutral evaluative suYxes that serve to create nouns with

an evaluative meaning component, as illustrated by the following examples

(Scalise 1988: 233):

(13) albero ‘‘tree’’ alber-ino ‘‘little tree’’

giardino ‘‘garden’’ giardin-etto ‘‘nice little garden’’

libro ‘‘book’’ libr-one ‘‘big book’’

ragazzo ‘‘boy’’ ragazz-accio ‘‘bad boy’’

In the domain of adjectives we Wnd many aYxes that can be attached to

adjectives without changing their syntactic category, in order to modify

their meaning: weakening (English -ish as in reddish, oldish), negation or

reversal (English dis-, in-, un-, non-), or intensiWcation of the meaning of the

adjectival base, as in the following examples from Dutch:

(14) bang ‘‘afraid’’ doods-bang ‘‘lit. death-afraid, very frightened’’

mooi ‘‘beautiful’’ bloed-mooi ‘‘lit. blood-beautiful, very beautiful’’

sterk ‘‘strong’’ bere-sterk ‘‘lit. bear-strong, very strong’’

triest ‘‘sad’’ in-triest ‘‘lit, in-sad, very sad’’

Formation of the middle verbs mentioned above illustrates that deriv-

ational processes are used to create speciWc subclasses of verbs. Middle

verbs in English are intransitive verbs, whereas their base verbs are transi-

tive, that is, require a direct object to be present. Middle verb formation is

therefore a case of change in syntactic valency. With the latter term we refer

to the syntactic combinatorial properties of words. Whereas intransitive

verbs only require the presence of a subject, transitive verbs require the

presence of a second noun phrase that functions as direct object. In many

languages, causative verbs are created from base verbs that mention an

event or action. The causative verbs have a subject that mentions the causer

of the event or action. Here are two examples from Bolivian Quechua, in

which the causative suYx -chi is attached to the verbs llank’a ‘‘to work’’

and awa ‘‘to knit’’, and thus induces a change in valency of these verbs. The

derived causative verbs take a subject that mentions the causer, in addition

to the NPs required by the base verb. For instance, in sentence (15a) the

presence of the NP churi ‘‘my son’’ is required by the base verb llank’

‘‘work’’ (van de Kerke 1996: 195–6):

(15) a. churi-y-ta llank’a-chi-saq

son-1sg-acc work-caus-1sg.fut

‘‘I will make my son work’’

60 word-formation

b. warmi-wawa-y-wan chompa-ta awa-chi-saq

woman-child-1sg-com sweater-acc knit-caus-1sg.fut

‘‘I will make my daughter knit a sweater’’

The Semitic languages are well known for their system of creating related

verbal lexemes by combining a consonantal root with a speciWc pattern of

C and V positions, and a vowel melody. In addition, preWxes may be used.

Such a set of verbal lexemes with the same morphological pattern is called a

binyan (‘‘building’’, plural binyanim). The root qtl ‘‘to kill’’ as used in biblical

Hebrewhas the Wve active binyan forms shown in (16) (AronoV 1994: 124). In

addition to these Wve binyanim, there are two binyanim with a passive mean-

ing, the Pu’al as a passive variant of the Pi’el, and the Hof’al as the passive of

theHif’il. Each of these seven binyanim has a paradigm of inXectional forms,

part ofwhich, the 3sg.masc.perfparadigm, is illustrated in (16).As theglosses

indicate, the binyan system mainly serves to create related verbs with diVer-

ences in syntactic valency.However, the semantics of the diVerent binyanim is

more complicated and less transparent than this example suggests.

(16) Binyan name 3sg.masc.perf Gloss

Qal qatal to kill

Nif’al ni-qtal to kill oneself

Pi’el qittel to massacre

Hif’il hi-qtil to cause to kill

Hitpa’el hit-qattel to kill oneself

3.2 Templates and idiosyncrasies

As discussed in section 1.4, complex words are listed in the lexicon for two

reasons. In the case of a fully regular complex word, it may be necessary to

list it because it is the conventional way of expressing a particular meaning.

For instance, the Dutch action noun for the verb concretiseer ‘‘to make

concrete’’ is concretiser-ing; the alternative noun concretis-atie is also

well-formed, but not the word that Dutch speakers in the Netherlands

use. Thus, established complex words may have a blocking eVect on the

coinage of new complex words. The second reason why we need to list

complex words is that they may have properties that are not predictable by

rule. Thus, morphological rules or templates have two functions: stating

the predictable properties of established complex words, and indicating

how new complex words can be formed.

derivation 61

This raises the issue of how to formally express the relation between the

regularities as stated in a rule or template, and the listed properties of

individual complex words. Consider the following German deverbal adjec-

tives in -bar ‘‘-able’’, and their idiosyncratic properties (Riehemann 1998: 54):

(17) additional aspect of meaning:

essen ‘‘eat’’ essbar ‘‘safely edible’’

obligation instead of possibility:

zahlen ‘‘pay’’ zahlbar ‘‘payable, must be payed’’

lexicalized meaning:

halten ‘‘hold, keep’’ haltbar ‘‘non-perishable’’

Fully regular adjectives in -bar are derived from transitive verbs, and their

meaning is predictable. This meaning can be circumscribed as ‘‘can be

V- ed’’. The three adjectives in (17) are derived from transitive verbs as

well, but they have an additional meaning or slighly diVerent meaning. For

instance, essbar does not just mean ‘‘can be eaten’’, but ‘‘can be eaten

safely’’. The meaning component ‘‘safely’’ is the additional unpredictable

meaning component. The adjective zahlbar does not mean that something

can be paid, but that it has to be paid. The meaning ‘‘non-perishable’’ of

the adjective haltbar is more speciWc than just ‘‘can be kept’’.

The relation between the properties that hold for all deverbal adjectives

ending in -bar, and the properties of the individual adjectives can be expressed

by an inheritance tree. In such a tree, each node inherits all the properties of

the dominating nodes, andmay have additional unique properties, which are

then also speciWed on that node. The following (simpliWed) inheritance tree

may be constructed for these German -bar adjectives (Figure 3.2). The unpre-

dictable properties of the three adjectives discussed are printed in italics. All

other properties of these three adjectives are predictable, and are inherited

from the dominating nodes of the base verbs and the dominating node with

Fig. 3.2 Inheritance tree for -bar adjectives

62 word-formation

the template for regular -bar adjectives. That is, these properties count as

redundant information. The predictable properties are partially inherited

from their base verbs, and partially from the word-formation template for

adjectives ending in -bar. The inheritance tree shown here is a multiple

inheritance tree since the lower nodes for the individual words are connected

to two diVerent layers of dominating nodes, the nodes of the base verbs, and

the node of the morphological template of -bar adjectives.

This kind of modelling of the lexicon enables us to specify which complex

words exist, and which of their properties have been generalized into

patterns, expressed as morphological templates. A morphological template

thus coexists with the individual complex words formed according to that

template. The templates will also be used to create new complex words of

that type which in their turn will be added to the lexicon. This model

reXects the way in which language users acquire the morphological system

of a language. First, individual complex words are learnt, and subse-

quently, more abstract patterns can be discovered on the basis of form-

meaning correspondences between sets of words.

Instead of providing each individual adjective with a speciWcation of all

its properties, one might also restrict this speciWcation to the unpredictable

properties that are not inherited from the dominating nodes. In that case,

the inheritance tree is redundancy-free. For instance, the speciWcation of

the regular English adjective doable will then consist of a node without any

speciWcation, linked to the nodes for to do and the relevant word-formation

template (Figure 3.3). In this case, the empty node at the bottom only

speciWes that the adjective doable is an existing adjective of English. All

other properties can be derived from the dominating nodes.

From a psychological point of view the full speciWcation approach is

more plausible since human memory is so vast that a massive amount of

redundant information is memorized, even though that information is

predictable. It is important to bear in mind that the existence of a particular

rule in a language does not imply that the outputs of such a rule cannot be

stored individually in the lexicon of that language. Note, Wnally, that this

Fig. 3.3 Non-redundant speciWcation of doable

derivation 63

point about the relation between rules and lists does not only pertain

to derivation, but also to compounding and inXection.

3.3 Constraints on derivation

Derivational processes impose constraints on the kind of base words they

take as their inputs. Examples of such input constraints are syntactic class

requirements, as illustrated above. Some derivational processes require

a more speciWc syntactic subclass. The Dutch suYx -baar ‘‘-able’’ can

only be used for deriving new adjectives from verbs that are transitive, as

in drink-baar ‘‘drinkable’’ derived from the transitive verb drink ‘‘to drink’’.

Does an aYx always require one speciWc syntactic category for its input

words? As we saw above, this is not the case for the Dutch negative preWx

on- that can be attached to adjectives and nouns. The Italian evaluative

suYxes mentioned in section 3.1 can be added to nouns, adjectives, and

adverbs. Even category-determining suYxes sometimes attach to words of

more than one category. The Dutch diminutive suYx attaches mostly to

nouns, but other categories are not excluded (section 3.1). The English

nominalizing suYx -er combines with both verbs and nouns (work-er,

London-er).

Input constraints may also be phonological in nature. The Dutch nom-

inalizing suYx -aar can only be attached to stems ending in a coronal

sonorant consonant preceded by a schwa. In other cases, the suYx -er

has to be used:

(18) luist[

e

]r ‘‘to listen’’ luister-aar ‘‘listener’’

duik[

e

]l ‘‘to tumble’’ duikel-aar ‘‘tumbler’’

bez[

e

]m ‘‘to sweep’’ bezem-er ‘‘sweeper’’

veeg ‘‘to sweep’’ veg-er ‘‘sweeper’’

The German diminutive suYx -lein is not attached to nouns ending in

/l/; instead, the synonymous suYx -chen is used. In Dutch, the suYx -ig

/

eg/ ‘‘-ish’’ is not added to words ending in /g/; instead, the synonymous

suYx -achtig is used. For instance, the denominal adjective *berg-ig

‘‘mountainous’’ is ill-formed, whereas berg-achtig is Wne. These German

and Dutch examples illustrate the cross-linguistic tendency to avoid

sequences with (almost) adjacent identical sounds in the phonetic forms

of complex words.

64 word-formation

The English verbalizing suYx -en that is attached to adjectives must be

preceded by exactly one stem-Wnal obstruent. This excludes a verb

*greenen, but allows for blacken, whiten, harden, fasten, and soften. In

the latter two examples, the /t/ is deleted in order to comply with this

phonological constraint. The constraint is an output constraint, because

the input soft ends in two obstruents, instead of one. Yet, it does not mean

that soft cannot be an input word; instead, its phonetic form is adjusted.

Thus, this verb is pronounced as [sOf e

n].

In sum, the possibility of using an aYx and the choice of a particular aYx

may be determined by considerations concerning the optimal phonetic

shape of the complex word that is being created. That is, output constraints

on the phonetic form of a complex word can govern the choice or usability

of a particular aYx, as these examples from German, Dutch, and English

illustrate.

Prosodic restrictions may also play a role in restricting productivity.

The term prosody refers to properties of words and sentences, such as

stress, tone, intonation, and the organization of sounds in larger units

such as syllables. These properties are not linked to speciWc segments,

and hence they are called suprasegmental properties.

An example of a prosodic constraint is the following. It has been obser-

ved for Dutch that a word almost never begins with two unstressed

preWxes. A stressed preWx, however, can easily be attached before an

unstressed one. Nominalization of preWxed verbs by means of the stressless

preWx ge- /g e

/ tends to be avoided, whereas the addition of the stressed

category-neutral preWx her- /hEr/ ‘‘re-’’ is without problems:

(19) ?ge-ont-dek ‘‘discovery’’, ?ge-ont-bos ‘‘deforestation’’, ?ge-be-loof

‘‘promising’’

her-ont-dek ‘‘to rediscover’’, her-be-gin ‘‘to re-begin’’, her-ge-bruik

‘‘to re-use’’

This diVerence in behaviour between the two preWxes has to do with the

tendency of Dutch to avoid sequences of unstressed syllables, in particular

at the beginning of a word. Therefore, the preWx ge- cannot be used before

another unstressed preWx.

Derivational processes may also be subject to stratal constraints. A

stratum is a layer of the lexicon of a particular historical origin. Germanic

languages have borrowed many words from Greek and Latin, often with

French as an intermediary language, words that form the non-native

derivation 65

stratum of the vocabulary. Since language users are able to discover aYxes

on the basis of recurrent form–meaning patterns in sets of borrowed words,

these languages acquired a set of non-native aYxes. These aYxes have not

been borrowed directly. It is words that are borrowed, and sets of borrowed

non-native words may give rise to non-native suYxes. Thus, the non-native

suYx -ity emerged in English, like its etymological cognates -iteit and -itat

in Dutch and German respectively. This suYx can now be used for coining

newde-adjectival nouns, forwhich there is no counterpart inLatin orFrench.

However, the use of such non-native aYxes (that is, neo-classical word-

formation) is mostly restricted to base words of the non-native stratum. The

Dutch suYx -iteit ‘‘-ity’’ can be attached to non-native adjectives, but not

to native ones. For native adjectives, the synonymous native suYx -heid

‘‘-ness’’ (of Germanic origin) has to be used. These native aYxes combine in

principle both with native and non-native stems. So the following asymmetry

can be observed:

(20) Non-native adjectives Derived nouns

absurd ‘‘absurd’’ absurd-iteit/absurd-heid ‘‘absurdity’’

stabiel ‘‘stable’’ stabil-iteit/stabiel-heid ‘‘stability’’

native adjectives

groen ‘‘green’’ *groen-iteit/groen-heid ‘‘green-ness’’

zeker ‘‘certain’’ *zeker-iteit/zeker-heid ‘‘certainty’’

One may wonder how language users, most of whom have no knowledge

of the historical origin of the words of their mother tongue, are able to

comply with such constraints. A feature such as [-native] cannot be seen

as an etymological feature, but only as a non-historical property of words.

A possible answer to this question is that non-native words tend to diVer

systematically from native ones in terms of their phonological make-up,

and thus they are recognizable. Dutch non-native adjectives, for instance,

are always polysyllabic and end in a stressed syllable (absurd ‘‘absurd’’,

stabıel ‘‘stable’’). This makes them diVerent from native adjectives that

are either monosyllabic (groen ‘‘green’’) or end in an unstressed syllable

(zeker ‘‘certain’’).

The choice of a particular aYx may also be determined by the morpho-

logical make-up of the base. For instance, when we want to nominalize an

English verb ending in -ize, the obvious choice is the suYx -ation, and not

some other suYx such as -ion (verbalization, not *verbaliz-ion). Adjectives

in -able prefer the suYx -ity to -ness, hence parsability, not parsableness.

66 word-formation

These are base-driven restrictions since it is the base that imposes

constraints on the suYx to be attached.

Such base-driven restrictions show that the internal morphological

structure of a complex word is accessible to morphological rules, whereas

the internal structure of complex words is not accessible for rules of syntax

(the Lexical Integrity Principle). Generally, the application of morpho-

logical rules may be sensitive to morphological structure. For example,

conversion of nouns to verbs in English and Dutch applies to simplex

nouns and compounds, but not to suYxed nouns.

An example of a semantic constraint is that the English suYx un- when

applied to verbs, indicates the reversal of a situation, as in unfold and

unscrew. Because of this meaning of un-, you cannot unswim or unkill.

Or, as the teacher warned my neighbour when he was taught how to

make proper holes in a Xute, ‘‘You can’t undrill a hole’’. This constraint

should not be interpreted as a constraint on a speciWc word-formation

process. The impossibility of unkill and similar verbs follows from the

incompatibility between the meaning of the deverbal suYx un- and that

of the base verb. In fact, there are possible worlds in which kill is not an

irreversible action, and where unkill could be used. That is, un- imposes

an interpretation of reversibility on the verb. This kind of semantic eVect is

called type coercion: the preWx forces the base verb to belong to a particular

semantic type of verbs, and thus imposes a particular interpretation on the

preWxed verb as a whole.

Derivational processes may also be subject to particular pragmatic,

stylistic, or sociolinguistic constraints. The use of the diminutive suYx -ie

in Dutch, as in the phrase lekker soepie ‘‘nice soup’’ and in liev-ie ‘‘love,

darling’’ is characteristic of an informal register, or is used between lovers

or in mother–child conversation. A more appropriate interpretation of

these pragmatic limitations on certain derivational patterns is to consider

them as linguistic means of deWning the communicative situation.

3.4 Productivity

The notion productivity is used quite frequently in morphological descrip-

tions of languages. It presupposes the idea of rule-governed morphological

creativity, the ability of language users to create new well-formed complex

words. Although this chapter deals with derivation, you should realize that

derivation 67

the notion ‘productivity’ is also relevant in the domains of compounding

and inXection. When we call a morphological pattern productive, we mean

that this pattern can be extended to new cases, can be used to form new

words. When we say that a morphological pattern is unproductive, this

means that it is not used for coining new words. The formation of plural

nouns by means of vowel change, as in English foot–feet has become

obsolete, and is an example of unproductive morphology. The same applies

to the use of vowel alternations for past tense verb formation in the

Germanic languages. You have to learn which verbs exhibit which particu-

lar vowel alternation, and the pattern is almost never used for making the

past tense of new verbs.

Within the class of productive morphological templates, we Wnd diVer-

ences in degree of productivity. Morphological patterns are not used to the

same degree. The basic opposition presupposed in using the notion ‘degree

of productivity’ is that between possible words and actual words. Leaving

inXection aside for a moment, one might say that the word-formation

templates of a language deWne the set of possible complex words of that

language. Not all these possible words are also actual words, that is, belong

to the lexical norm or convention of that language. For instance, my English

dictionaries do not mention the verb unblacken although it is a possible

word with a transparent meaning, ‘‘undoing the eVect of blackening’’.

The degree of productivity of a word-formation pattern thus refers to the

degree to which the structural possibilities of a word-formation pattern are

actually used. This can be illustrated as follows. Both in German and in

Dutch female nouns can be coined by means of suYxation, in German by

means of suYxation with -in:

(21) Dozent ‘‘teacher’’ Dozent-in ‘‘female teacher’’

Minister ‘‘minister’’ Minister-in ‘‘female minister’’

Professor ‘‘professor’’ Professor-in ‘‘female professor’’

Student ‘‘student’’ Student-in ‘‘female student’’

Dutch has the same female suYx -in, and some other female suYxes as well.

Yet, it does not have equivalent words for these German female nouns.

A German speaker may begin his speech with the phrase Liebe Horerinnen

und Horer ‘‘dear female listeners and listeners’’ whereas Dutch speakers

will never say the equivalent, well-formed expression Geachte toehoorsters

en toehoorders. This shows that German requires more awareness of there

being a sex diVerence between the people addressed than Dutch that does

68 word-formation

not exploit this word-formation possibility very much. So you see it makes

sense to say that certain morphological processes are more productive than

others. The actual use of structural possibilities may have to do with

language-external factors such as cultural habits and politeness rules.

In addition, there are language-internal factors such as the existence of

competing word-formation processes.

Morphological processes can lose their productivity in the course of

time. The Dutch suYx -lijk ‘‘-able’’ used to be productive in earlier stages

of Dutch, but has now become completely unproductive, and its role has

been taken over by the synonymous suYx -baar. So the set of Dutch

adjectives ending in -lijk has become a closed set of words that cannot be

extended any more.

How can we measure the degree of productivity of a morphological

pattern, and how can we rank the diVerent patterns on a scale of product-

ivity? We might count the number of diVerent words (word types) of a

certain morphological type, i.e. its type frequency, to be distinguished from

the notion token frequency. The token frequency of a morphological class of

words is the summed frequency of use of all the words of that particular type

in a sample of language use. For example, if there are 100 diVerent English

adjectives in -able in a certain language sample, the type frequency of -able

adjectives is 100, but the token frequency will be much higher because

the individual adjectives in -able will have been used more than once.

Type frequency is best calculated on the basis of a large corpus instead of a

dictionary. A corpus is a collection of data concerning actual language use,

these days mainly in electronic form. A corpus is a better source of informa-

tion than a dictionary. A dictionary is always lagging behind with respect to

the use of productive morphological patterns because it only registers, after

some lapse of time, which new complex words have become established

words. Morphological productivity manifests itself most clearly in the

appearance of complex words that never make it to the dictionary. However,

a high type frequency of a morphological pattern, even when based on a

proper corpus, does not tell us that much about degree of productivity: the

class of words of that type may be a more or less closed set that is (almost)

never expanded. Therefore, we need another way of computing productivity.

A basic property of a productive word-formation process is that it may

lead to hapaxes, new word types that occur only once in the corpus, and

clearly do not belong to the set of established words. Therefore, one might

deWne the degree of productivity P of a particular morphological process as

derivation 69

the proportion between the number of hapaxes of that type (n1) to the total

number of tokens N of complex words of that type in the sample (Baayen

1992: 115):

(22) P ¼ n1=N

The use of P as a measure of productivity is illustrated by the data in

Table 3.1. These data are from the English Cobuild Corpus, a corpus of

18 million word forms of British English. N stands for the number of word

tokens ending in these aYxes, andV for the number of types. The table shows

that the number of tokens in -ity is higher than that of the tokens in -ness.

However, the number of types with -ness is higher, and—what is more

important—the number of hapaxes as well. Hence, the suYx -ness is

more productive than -ity. This is in accordance with the observation that

the use of -ity is restricted to being attached to non-native stems, whereas

-ness can beusedwith all kinds of stems.Whereas both stability and stableness

are well-formed, we do not have a well-formed word reddity besides redness.

The data in this table concern two rival aYxeswith the samemeaning. Thus

it is possible to compare them because they have the same potential of being

useful for the language user, and to rank them on a scale of productivity.

However, it may be the case that an aYx has a relatively high proportion of

hapaxes in the set of token words with that aYx, without that aYx creating a

lot of newwords, simply because that kindof complexword is not very useful,

has no high pragmatic potential. If we want to get a more precise idea of the

contribution of a particular type of complex word to the growth of the

lexicon, we might use another measure, referred to as the global productivity

measure P*, which is the number of hapaxes of that morphological type in a

given corpus divided by the total number of hapax words in that corpus.

Thismeasure gives us a ranking for all aYxes on one scale, and indicates their

relative contribution to the growth of the lexicon. Thus, we have at least

three measures for quantifying three diVerent aspects of productivity: the

Table 3.1. Productivity measure for the English suYxes -ity and -ness

AYx N V n1 P

-ity 42,252 405 29 0.0007

-ness 17,481 497 77 0.0044

Source: Baayen 1992: 116.

70 word-formation

number of attested types (V), the potential to make new word types (P), and

the actual contribution to the growth of the lexicon (P*).

The existence of rival aYxes may also aVect the token frequency of the

corresponding types of complex word. Once a complex word with one of

these aYxes has been formed and established, the language user will tend

not to use a rival process for creating another complex word with the same

meaning. This is called the blocking eVect. However, if the rival processes

are very productive, blocking is not a very strong factor. As illustrated by

the examples (20), pairs of complex words with exactly the same meaning

are possible. In other cases, complex words with the same base exhibit

semantic diVerences. For instance, there is a semantic diVerence between

admission and admittance, both derived from the verb to admit. If there is

such a diVerence, it is obvious that there cannot be a blocking eVect.

3.5 Affix ordering

Complex words may contain more than one preWx or suYx, and we would

therefore like to know which principles govern the order of aYxes. Given

the distinction between derivation and inXection (derivation creates lex-

emes, inXection creates forms of lexemes), we expect the following schema

to apply, and this is indeed basically correct (although there are some

complications, see Chapter 5):

(23) InXectional preWxes—Derivational preWxes—Root—Derivational suYxes—

InXectional SuYxes

As to ordering restrictions within a sequence of derivational aYxes, it is

clear that the input constraints often predict which order is the correct one.

For instance, the suYx order in read-abil-ity is determined by the fact that

the suYx -ity selects adjectives as bases and creates nouns, whereas -able

takes verbs as inputs in order to form adjectives. Therefore, -able must be

attached before -ity.

Stratal restrictions account for the generalization that in Germanic lan-

guages native suYxes are peripheral with respect to non-native ones. As we

saw, it is only non-native suYxes that require their inputs to be of a speciWc

stratum, the non-native one. Native suYxes combine with both non-native

and native inputs.Once a native suYx has been added to a stem, the newword

is [+ native], which blocks the attachment of non-native suYxes. Therefore,

derivation 71

the only possible order is: non-native root + non-native suYxes + native

suYxes. For instance, theDutch native suYx -baar ‘‘-able’’ can be attached to

verbal stems ending in the non-native suYx -iseer (as in stabiliseer-baar

‘‘stabilizable’’), but not vice versa: *drag-bar-iseer ‘‘port-ableize’’ is ill-formed

because the non-native suYx -iseer has been added to a native stem.

AYx order might also be determined by the intended semantic scope of

the aYxes used. Thus, derivational aYxes may appear in diVerent order,

with diVerent interpretational eVects, as illustrated here for Quechua which

allows for both the suYx sequence -schi-rpari- and the suYx sequence

-rpari-schi- (Muysken 1988: 267–8):

(24) a. llank’a-schi-rpari-n

work-help-eVect-3sg

‘‘He really helped him work’’

b. qunqa-rpari-schi-wa-n

forget-eVect-help-lobj-3sg

‘‘He helped me forget completely’’

The ordering of, in particular inXectional, aYxes is sometimes described in

terms of templates or position classes (Chapter 5).

Summary

Derivation is the formation of lexemes by means of aYxation, conversion,

reduplication, and root-and-pattern morphology. These diVerent morpho-

logical operations serve a large number of semantic and syntactic functions,

including that of valency change. Such derived words, once they are formed,

may exhibit idiosyncratic properties. Word-formation templates function

as redundancy statements with respect to established, listed complex

words, which can be modelled by inheritance trees.

The application of a particular word-formation process may be governed

by rule-speciWc andgeneral input constraintswith respect to syntactic, phono-

logical, morphological, stratal, and semantic properties of the base words.

Morphological rules diVer in their degree of quantitative productivity,

which can be estimated by counting the number of hapaxes of the relevant

morphological type in a sample of words.

An important topic of morphological analysis is the investigation of

principles and constraints with respect to the ordering of aYxes in a

multiply complex word.

72 word-formation

Questions

1. The English verb to ride has as its past tense form rode, whereas the past tenseof the denominal verb to joyride is joyrided. Suppose that the use of vowelchange for making past tense forms is triggered by the presence of a diacriticfeature [+ ablaut]. How can the difference between rode and joyrided beexplained?

2. The English suffix -able is of non-native, Romance origin. Can you provideevidence to prove that this suffix has become a native suffix in present-dayEnglish?

3. Do the following English denominal verbs form a problem for the Right-hand Head Rule for English: enamour, encage, enchain, encircle, encourage,enfeeble? (Bear in mind that possible words can form an intermediate stagein word-formation.)

4. The Italian default conjugation is that of verbs in -are, as can be seen inItalian loan verbs from English such as dribblare ‘‘to dribble’’ and scioccare‘‘to shock’’. How can this fact be used for interpreting the formation of theItalian verb invecchiare in (6) as a case of prefixation only instead of con-sidering it a case of parasynthetic word-formation?

5. A number of Dutch prefixed verbs have a corresponding noun withoutadditional overt morphology. Such nouns, derived from prefixed verbsthrough conversion, are always neuter:

be-roep ‘‘to appeal’’ beroep ‘‘appeal’’ge-bruik ‘‘to use’’ gebruik ‘‘use’’onder-wijs ‘‘to teach’’ onderwijs ‘‘teaching’’over-leg ‘‘to deliberate’’ overleg ‘‘deliberation’’ver-val ‘‘to decay’’ verval ‘‘decay’’ver-raad ‘‘to betray’’ verraad ‘‘betrayal’’

Nouns derived from simplex verbs through conversion (such as val ‘‘fall’’),however, tend to be non-neuter. Why are these facts a problem for apercolation account of gender assignment to complex nouns?

6. In Malagasy, reduplication is often used for attenuation or diminution. Thus,fotsifotsy ‘‘white-white’’ means ‘‘whitish’’ (compare the simplex word fotsy‘‘white’’; KeenanandPolinsky1998: 571).Doyou think it is possible toexplainthis interpretation of reduplicated forms bymeans of the notion of ‘iconicity’?

7. English has at least the following prefixes with some negative meaning: de-,dis-, in-, non-, and un-. Make a list of the (phonological, syntactic, semantic,and stratal) constraints that each of these prefixes imposes on its basewords. What is the division of labour between these five prefixes?

8. a. In Polish, the diminutive suffix -awy can be attached to adjectivesonly, resulting in adjectives with the meaning ‘‘somewhat A’’, as inczarny-czarnawy ‘‘black–blackish’’ and czerwony–czerwonawy ‘‘red–reddish’’. Is this a counterexample to the RHR if that rule is assumedto be valid for Polish?

derivation 73

b. This diminutive suffix cannot be attached to relational adjectives suchas domowy ‘‘of a house’’ derived from dom ‘‘house’’, or szkolny ‘‘of aschool’’ derived from szkoła ‘‘school’’. How might this be explained?

9. The English negative prefixes a- and an- borrowed from Greek can both beattached to adjectives. What determines the choice between these prefixes?

10. The relational adjective American can be modified by the adverb very, as ina very American attitude, or prefixed, as in an un-American attitude. Whatkind of change in the interpretation of American does this bring about?

Further reading

General discussions of constraints on word-formation are found in Aronoff(1976), Booij (1977, 2002b), Plank (1981), Plag (1999, 2003), Scalise(1984), and Rainer (2000). Plag (1999) advocates base-driven restrictions.Anderson (1992) defends the position that word-internal structure is inaccess-ible to morphology (the theory of A-morphous Morphology), whereasCarstairs-McCarthy (1993) defends the opposite view. A survey of cases ofcross-categorial morphology is given in Plank (1981: 43–65).The binyanim of Classical Hebrew are treated in Aronoff (1994) and Verheij

(2000).Bauer (2001) presents a survey and analysis of the discussions on the notion

‘productivity’; quantitative measures of productivity are proposed in Baayen(1992, 1993). The different quantitative measures of productivity are alsodiscussed in Plag (1999: ch 2) and in Baayen (2003).Affix ordering is discussed in the framework of lexical phonology, in terms of

so-called level ordering (cf. Fabb 1988; Plag 1996; Booij 2000). Affix orderingprinciples for English and German are discussed in Aronoff and Fuhrhop(2002), and in a number of articles in YoM 2001. Hay (2002) relates affixordering restrictions in English to the degree of parsability of the affixes. Rice(2000) proposes semantic principles for affix order in Athapaskan languages.

74 word-formation

4Compounding

4.1 Compound types

In many languages, compounding (also called composition) is the most

frequently used way of making new lexemes. Its deWning property is that

it consists of the combination of lexemes into larger words. In simple cases,

compounding consists of the combination of two words, in which one word

modiWes the meaning of the other, the head. This means that such com-

pounds have a binary structure. Here are some examples from a number of

European languages:

(1) Dutch huis-vrouw ‘‘house wife’’

German Rot-licht ‘‘red light’’

Greek organo-pektis ‘‘instrument player, musician’’

Hungarian asztal-la2b ‘‘table-leg’’

Latin perenni-servus ‘‘perennial slave, slave forever’’

The productivity of compounding in many languages is largely due to its

semantic transparency and versatility. When a new compound is formed,

we already know the meaning of its constituents, and the only task we face

is to Wnd out about the semantic relation between the two parts. The

general semantic pattern of a compound of the form XY is that it denotes

a Y that has something to do with X, or vice versa, depending on the

language. The exact nature of the semantic relation between the two

4.1 Compound types 75

4.2 Compounds and phrases 81

4.3 Compounds and derived

words 85

4.4 InterWxes and allomorphy 88

4.5 Synthetic compounds

and noun incorporation 90

Summary 93

Questions 93

Further reading 95

constituents receives no formal expression, and is a matter of interpretation

by the language user. As language users, we have to interpret that relation-

ship on the basis of the meanings of the compound constituents, our

knowledge of the world, and sometimes the context in which the compound

is used. The role of context can be illustrated by a situation in which we

tell a guest who enters the room and looks for a seat: You may take the

apple-juice chair. This means that this person should take the seat at the

table in front of which a glass of apple juice is standing (Downing 1977).

The recently coined compound butt call denotes a call on your mobile

phone from someone who calls you unintentionally by sitting on her/

his own mobile phone and thus pressing a button. This is by nature a

contextual compound, but even these can get lexicalized.

The process of compounding can be applied recursively, and thus we

might get pretty long compounds. This is another cause of its productivity.

A famous long compound is the German word Donau-dampf-schiV-fahrts-

gesellschaft ‘‘lit. Danube-steam-ship-travel-company’’, the name of a ship-

ping company that used to be active on the Danube; an English example is

White House travel oYce staV. The structure of this latter compound can be

represented as follows:

(2) [[[[White]A [House]N]N [[travel]N [oYce]N]N]N [staV ]N]N

Instead of labelled bracketing we may also use trees to represent the

morphological structure of compounds. In the case of multiply complex

words, this notation is easier for understanding the structure of a word.

The tree representation of the compound in (2) is as shown in Figure 4.1.

The morphological notion ‘head’ discussed in Chapter 3 is very relevant

for the analysis of compounding. In Germanic languages like English, for

instance, the right constituent of a compound is normally the head, and

hence almost all compounds conform to the following scheme:

Fig. 4.1 Morphological tree of White House travel oYce staV

76 word-formation

(3) [ X Y]Y, Y ¼ N, A, V

As shown by words such as underdog and German Um-welt ‘‘lit. around-

world, environment’’ (um ‘‘around’’ is a preposition), not only content

words may occupy the non-head position X, but function words such as

prepositions as well. As we will see shortly, phrases may also occur in

the non-head position. Almost all compounds in these languages belong

to one of the major lexical categories, in particular N and A. Compounds

with a verbal head (Y ¼ V) do occur, but are exceptional in Germanic

languages.

The scheme in (3) indicates that the syntactic category of the compound

as a whole is that of the right constituent. This also holds for subclass

features, such as gender for nouns. In Dutch, the distinction between

neuter and non-neuter gender manifests itself in the choice of the def.sg

determiner (het and de respectively), and so we get pairs like the following:

(4) het soepvlees ‘‘the soup meat’’

(neuter)

de vleessoep ‘‘the meat soup’’

(non-neuter)

het modefeest ‘‘the fashion party’’

(neuter)

de feestmode ‘‘the party fashion’’

(non-neuter)

The headedness of a compound is not only relevant for its formal

properties, but also for its semantic interpretation. The compound soepvlees

in (4), for example, denotes a particular kind of meat, not a particular kind

of soup, whereas the inverse applies to the compound vleessoep. In other

words, the left constituent of such compounds functions to modify the

meaning of the head constituent.

The notion ‘head’ is also relevant for the application of inXectional rules,

since it is the head that determines how the inXectional properties of the

whole compound are realized. Since the Dutch noun kind ‘‘child’’ has the

irregular plural form kinderen, this plural form recurs in the plural form of

the compound kleinkind ‘‘grandchild’’, namely kleinkinderen. This is why

the process of noun pluralization is sometimes called a head operation. An

alternative interpretation of these facts is that the pluralization of a com-

pound is determined by its paradigmatic relationship with its head noun. In

this example, the compound kleinkind is then paradigmatically related to

the noun pair kind–kinderen ‘‘child–children’’, and hence its plural is klein-

kinderen.

Compounds are not universally right-headed since there are also

languages with left-headed compounds. The following examples of

compounding 77

left-headed compounds are from Maori, spoken in New Zealand

(Bauer 1993: 518–20):

(5) roro hiko

brain electricity

‘‘computer’’

maarama taka

month revolve

‘‘calendar’’

wai mangu

water black

‘‘ink’’

whare heihei

house hen

‘‘hen-house’’

This diVerence with respect to the position of the head in compounds might

suggest that the position of the head is a parametrical diVerence between

languages. For Germanic languages, the value of the parameter is ‘right’,

whereas it is ‘left’ for Maori.

Is it possible for a language to have both left-headed and right-headed

compounds? Italian appears to have both left-headed compounds such as

capo-stazione ‘‘lit. master station, station master’’ and croce-rossa ‘‘lit.

cross red, red cross’’, and right-headed compounds such as gentil-uomo

‘‘kind man, gentleman’’ (Scalise 1992a). However, one could also interpret

the left-headed compounds as lexicalized phrases. This interpretation

is supported by the observation that the plural form of capostazione is

capistazione, with an internal plural suYx -i. So the answer to the question

raised in this paragraph depends on your criteria for word-hood. If we

want to stick to the rule that inXection cannot appear word-internally,

capostazione is a phrase (and thus Italian might be a language with right-

headed compounds only), but if we think this is not a relevant criterion, it

can be classiWed as a (left-headed) compound with inXection on its head

capo, that is, a case of head inXection.

The scheme in (3) above is a bit too general for Germanic languages such

as English, in that it predicts too many compounds to occur. A robust

generalization for Germanic languages is that VV compounds such as to

freeze-dry, AV compounds such as to whitewash, and NV compounds such

as to machine-wash are pretty rare. They have often been coined through

78 word-formation

back formation from nominal compounds (to babysit from babysitter) or

adjectival compounds (to machine-wash frommachine-washable). Hence we

have to specify which instantiations of (3) are well formed in a particular

language, and which can be formed productively.

Another source of compound verbs is the conversion of nominal com-

pounds into verbs. For instance, the Dutch nominal compound voetbal ‘‘soc-

cer’’ has been converted into the verb voetbal ‘‘to play soccer’’. Note that this

verb has no right verbal constituent as its head. Therefore, we have to specify

for each language with right-headed compounds which particular combin-

ations of lexical categories in scheme (3) can be used for coining compounds.

The left, non-head position X of scheme (3) allows for all kinds of

constituents, and hence is left unspeciWed. Phrasal constituents need not

be excluded from the left head position, as the following examples from

English illustrate:

(6) [French history]NP teacher

[20th century]NP welfare state

[wages or employment]NP protection

[‘one size Wts all’]S mass production public service

It is obvious why phrases can only occur in the left, non-head position: if

they appeared in the head position, such constructions would be phrases

themselves, and not words. However, the possibility for phrases to appear

within compounds does not mean that all kinds of phrases are allowed in

this position. DeWnite NPs with a determiner are clearly excluded, as illus-

trated by the ungrammaticality of the phrase a [the French history] teacher.

Compounds with a head are called endocentric compounds. The term

‘endocentric’ means that the category of the whole (syntactic or morpho-

logical) construction is identical to that of one of its constituents. There are

also exocentric compounds for which this is not the case. Consider Romance

compounds consisting of a verbal stem followed by a plural noun (Scalise

1992a; Rainer and Varela 1992):

(7) Italian porta-lettere ‘‘lit. carry letters, postman’’

lava-piatti ‘‘lit. wash dishes, dish washer’’

Spanish lanza-cohetes ‘‘lit. launch rockets, rocket launcher’’

limpia-botas ‘‘lit. clean boots, bootblack’’

These compounds are nouns, but there is no head. In this respect, they are

comparable to English compounds like pickpocket and cut-throat that also

compounding 79

lack a head. The plural nominal constituents do not function as heads. For

instance, portalettere does not denote certain kinds of letters. The left

constituent is a verb in its stem-form. The nouns in the right position of

these compounds have plural endings, whereas the compounds themselves

are neutral with respect to number, and can be used both as singular and as

plural forms. Therefore, these nouns cannot be the heads of the compounds

of which they form a part. So this kind of compound is a clear case of

exocentric compounding.

A special semantic interpretation is required for compounds such as

German Kahlkopf ‘‘lit. bald-head, person with a bald head’’, and English

blue-stocking. In these cases, the compound denotes the person who is in

possession of the entity mentioned by the compound. The traditional term

for compounds with this special interpretation, taken from Sanskrit, is

bahuvrihi-compound. They are sometimes considered to form a subset of

the exocentric compounds, since they do not refer to the entity mentioned

by the head of the compound: a baldhead is not a type of head.However, this

special use is not unique for compounds. When a teacher wants to address a

pupil whose name she does not know, shemight say:That red sweater should

shut up, meaning that the pupil with the red sweater should keep his mouth

closed. Hence, this use of compounds appears to be a conventionalized use

of words, in which the word for a part denotes the whole (pars-pro-toto).

Therefore, we do not have to consider them as a special structural category.

The conventional interpretation might however inXuence its formal behav-

iour. The Dutch bahuvrihi-compound spleet-oog ‘‘slit-eye’’ is used to refer

to people of Chinese appearance. Its gender is non-neuter, although its head

oog ‘‘eye’’ is a neuter noun. So the formal gender class of the head noun is

overruled by the special semantic interpretation of this compound.

Another type of bahuvrihi-compounds is exempliWed by the Latin

adjectival compounds auri-com-us ‘‘having golden hair’’ and magn-anim-us

‘‘magnanimous’’. Here the noun aurum ‘‘gold’’ and the adjective magnus

‘‘great’’ combine with a noun (coma ‘‘hair’’ and animus ‘‘soul’’ respectively)

into a compound that is an adjective (Oniga 1992). The exocentricity of

these compounds cannot be explained in terms of semantic interpretation

in the same way as baldhead, because they behave formally as adjectives,

although there is no adjectival head.

A special class of compounds is formed by copulative compounds. In

these compounds there is no semantic head, and the relation between the

constituent is a relation of coordination. Examples of such compounds are

80 word-formation

the Sanskrit dvandva compounds, and similar compounds in Punjabi (Olsen

2001; Bhatia 1993: 320):

(8) candra-ditya-u

moon-sun-dual

‘‘the moon and the sun’’

deva-sura-s

god-demon-pl

‘‘three or more gods and demons’’

raat-din ‘‘night and day’’

maa-pio ‘‘mother and father’’

sukh-dukh ‘‘happiness and sorrow’’

These compounds function as dual or plural expressions, and are therefore

quite similar to NPs with coordination, which also receive a plural

interpretation.

Copulative compounds also occur in European languages, as illustrated

by the following words:

(9) German Osterreich-Ungarn ‘‘Austria and Hungary’’

Furstbischof ‘‘prince and bishop’’

English blue-green, washer-dryer

Dutch rood-wit-blauw ‘‘red-white-blue’’

The last example is an adjectival compound with three constituents of

semantically equal status. Yet, it is only the last of the three constituents

that receives the inXectional ending -e in attributive position (as in rood-wit-

blauw-e vlag ‘‘red-white-blue Xag’’), which shows that this is not a case of

syntactic coordination of words. If that were the case, all adjectives would

be inXected, as in *rode-witte-blauwe vlag. However, the copulative com-

pounds in (9) are diVerent from dvandva compounds because their number

is singular. A Furstbischof is a person who is simultaneously a prince and a

bishop, and this word does not refer to a combination of persons. Therefore,

such compounds are sometimes classiWed as appositive compounds.

4.2 Compounds and phrases

An important issue in the analysis of compounds is their demarcation from

phrasal expressions. There are two reasons why it is not always easy to

distinguish the two. First, phrases can have the same function as words, that

compounding 81

of labels for name-worthy categories. Second, phrases and compounds

look quite similar because compound patterns often derive historically

from phrasal word combinations.

Let us now have a look at formal correlates of the distinction between

compounds and phrases. The German word for ‘‘red cabbage’’ is the

Adjective–Noun compound Rotkohl, whereas Dutch uses the NP rode

kool to express the notion ‘‘red cabbage’’. In this example it is not diYcult

to see that the Dutch expression is a phrase, since the adjective is inXected

(the stem rod- ‘‘red’’ is followed by the inXectional ending -e). This type of

inXection is a case of agreement: the adjective has to agree with the noun that

it modiWes for properties such as gender, number, and deWniteness. The

German expression, on the other hand, is a compound because the adjective

Rot is not inXected (compare this to the German noun phrase ein rot-er Kohl

‘‘a red cabbage’’, with the adjectival ending -er). This reasoning presupposes

that word-internal constituents cannot be aVected by a syntactically

conditioned rule such as agreement. Independent evidence for the diVerent

status of German Rotkohl compared to Dutch rode kool is found in the

stress diVerences between these expressions. In German and Dutch noun

phrases, main stress is normally on the head, whereas in nominal com-

pounds main stress is on the non-head. Indeed, the main stress locations

are diVerent:Rotkohl (compound stress) versus rode kool (phrasal stress). In

English there is no adjectival inXection of this kind, and hence stress is the

only criterion that we can use to distinguish phrases from compounds.

Thus, we consider blackboard a compound, and black board a phrase.

This use of A +N phrases as labels for categories implies a certain formal

restriction: the adjective cannot bemodiWed. A phrase such as heel rode kool

‘‘very red cabbage’’ cannot function to classify kool ‘‘cabbage’’. This latter

phrase functions as a description instead of a name for a particular kind

of cabbage.

The potential functional equivalence of compounds and phrases of the

Adjective + Noun type is particularly clear in the case of noun phrases

with relational adjectives. Consider the following data from English

(Levi 1978: 38):

(10) a. atom bomb b. atomic bomb

industry output industrial output

language skills linguistic skills

city parks urban parks

ocean life marine life

82 word-formation

We see NN compounds in the left, and AN phrases in the right column.

The adjectives used are denominal relational adjectives. We consider

the expressions in the right column as phrases because they carry main

stress on their right constituent. The last two examples illustrate that these

relational adjectives do not necessarily have to be the adjectival derivatives

of nouns: urban, for instance, is the only available relational adjective for

expressing the meaning ‘‘related to cities’’. These relational adjectives can

normally only be used in attributive position. We can speak of urban parks

or rural police, but not of parks that are urban or police that is rural

(we might, however, speak of parks that are urban in nature). In sum,

AN phrases with relational adjectives are functionally equivalent to NN

compounds. Yet, these two constructions have to be kept apart as far

as their formal properties are concerned, as phrases versus compounds.

Another illustration of this demarcation problem are the so-called geni-

tive compounds in English of the type women’s magazine, girls’ school,

Down’s syndrome andMurphy’s law. These expressions must be considered

phrases given the presence of the internal inXectional suYx or clitic -s. Yet,

they function in the same way as compounds. These expressions instantiate

a lexicalized syntactic pattern that functions to create new labels. Note also

that many of these expressions have main stress on their Wrst constituent,

just like English compounds: Down’s syndrome and Murphy’s law. We

therefore assume an idiomatic pattern or constructional idiom N’s N for

English that serves to create new lexical expressions. A constructional

idiom is a Wxed syntactic pattern in which some positions may be Wlled by

all kinds of words of the right category, whereas other positions are Wlled

by speciWc morphemes or words. In this case, there is only one morpheme

lexically speciWed, the morpheme -s. The two N positions are variable, and

can be Wlled by all sorts of noun.

Certain lexical expressions in Romance languages are sometimes incor-

rectly called compounds although they have in fact a phrasal form. This

applies to French salle a manger ‘‘dining room’’ and chambre d’hote ‘‘guest

room’’. The structures N a N and N de N are instantiations of the syntactic

structure [N PP]NP, a noun phrase consisting of a head N followed by a PP

complement, and have developed into constructional idioms. Such phrases

are functionally equivalent to compounds in Germanic languages, and that

is why the mistake is made to consider them compounds. Note, however,

that their plural forms are salle-s a manger and chambre-s d’hotes respect-

ively, with an internal plural ending. This proves their phrasal nature since

compounding 83

the plural form of a French word is expressed by a suYx at its right edge.

Another type of apparent French compound is homme-grenouille ‘‘lit. man

frog, frogman’’. Its plural form requires both constituents to be pluralized

(hommes-grenouilles) which suggests that we have to dowith anNP inwhich

the phrasal head is followed by a noun with an appositional function.

The word-or-phrase problem also shows up in the formal analysis of

separable complex verbs. These are verbal expressions that look like verbal

compounds, but do not have the formal status of words since their con-

stituents can be separated in syntax. Consider the following Hungarian

noun–verb combinations (Kiefer 1992):

(11) level-et ır

letter-acc write

‘‘be engaged in letter writing’’

ujsag-ot olvas

newspaper-acc read

‘‘be engaged in newspaper reading’’

teve-t nez

television-acc watch

‘‘be engaged in television watching’’

Similar separable complex verbs occur in Dutch:

(12) stof zuigen

dust suck

‘‘to vacuum-clean’’

bier brouwen

beer brew

‘‘to brew beer’’

piano spelen

piano play

‘‘to play the piano’’

These Hungarian and Dutch expressions are diVerent from regular verb

phrases in that the object NP is a bare noun and cannot be preceded by a

determiner. This makes them look like verbal compounds. However, these

expressions are phrasal in that they can be split in certain syntactic con-

texts. In Hungarian, the negative particle nem ‘‘not’’ can appear between

the noun and the verb:

(13) Levelet nem ır ‘‘He is not engaged in letter writing’’

84 word-formation

In the case of the Dutch separable complex verbs, their phrasal nature can

be concluded from the fact that the participle preWx ge- does not appear

before the particle, but in between the particle and the verb. The relevant

participles of the verbs mentioned above are stof-ge-zog-en, bier-ge-brouw-

en and piano-ge-speel-d respectively.

Phrasal patterns with a word-like function such as these separable com-

plex verbs can be qualiWed as constructional idioms. The constructional

idiom ‘bare Noun + V’ in Hungarian and Dutch has the speciWc meaning

‘‘to be engaged in a particular institutionalized activity’’ (such as writing

letters or playing the piano). Such constructional idioms may serve the

same function as morphological patterns: expanding the set of lexical units

of a language. Recall that the concept of constructional idiom is also

applicable to the cases of apparent compounds in French discussed

above. These are syntactic patterns such as N de N, with open positions

for the nouns, and a Wxed preposition de, patterns that can be used to coin

new expressions to designate classes of entities.

4.3 Compounds and derived words

The crucial distinction between compounds and derived words is that in

compounds each of the constituents is a form of a lexeme, whereas deriv-

ation involves aYxes, that is, non-lexemic morphemes. However, the dis-

tinction is not always so clear-cut, because a lexeme may develop into a

derivational morpheme. An example is the Dutch noun boer ‘‘farmer’’ that

occurs in complex words such as the following:

(14) groente-boer ‘‘lit. greens farmer, green-grocer’’

melk-boer ‘‘lit. milk farmer, dairy man’’

sigaren-boer ‘‘lit. cigars farmer, cigar seller’’

tijdschriften-boer ‘‘lit. magazines farmer, magazine seller’’

In the Wrst two examples, the original meaning of ‘‘farmer’’ still makes

some sense since farmers may sell their produce such as greens and dairy.

However, these words are nowadays used to refer to persons who sell veget-

ables or dairy without producing these goods themselves. The last two

examples show even more clearly that the morpheme boer has developed

into a morpheme with the meaning ‘‘seller’’, but only in combination with

another noun. Hence, we may conclude that boer has developed into a

compounding 85

suYx. In fact, many aYxes derive from lexemes. An example of a preWx that

derives from a lexeme is Dutch oud ‘‘old’’ that has the meaning ‘‘former,

ex-’’ when added to a nounas in oud-burgemeester ‘‘ex-mayor’’.Note that this

word cannot receive the interpretation ‘‘old mayor’’. The phenomenon of

lexemes becoming aYxes is a cross-linguistically widespread phenomenon,

and an instance of grammaticalization, the historical process in which

lexical morphemes become grammatical ones. Grammatical morphemes are

either function words or bound morphemes. AYx-like morphemes such as

boer and oud that still correspond to a lexeme are called aYxoids.

The boundary between compounding and derivation is also blurred in

the domain of neo-classical compounding. In this kind of compounding

one or both of the constituents of a word are roots borrowed from Greek

and Latin that do not correspond to lexemes, so-called combining forms.

Consider the following English words:

(15) a. bio-logy, psycho-logy, socio-logy, geo-graphy, tomo-graphy

b. tele-camera, tele-graph, tele-gram, tele-kinesis, tele-matics, tele-phone,

tele-vision

c. bureau-crat, magneto-metry, magneto-hydro-dynamic

The examples in (15a) consist of two combining forms. Combining forms

are divided into initial combining forms (bio-, psycho-, socio-, geo-, tomo-),

and Wnal combining forms (-logy, -graphy). The root graph is found as a

Wnal combining form in telegraph in (15b). There is also a lexeme graph, a

word used in mathematics. However, the lexeme graph has a speciWc

mathematical meaning that does not show up in telegraph.

Neo-classical compounds are diVerent from normal compounds in that

the meanings of the constituent parts cannot be derived from the meaning

of corresponding lexemes. For instance, when a language user is able to

assign the meaning ‘‘life’’ to the morpheme bio-, this is either based on

comparison of a number of words that begin with bio- (in the same way in

which we discover the meaning of aYxes), or because she has been

taught this at school, maybe since she took Ancient Greek as a subject.

The reason why we call such words neo-classical compounds is that in

most cases they have not been borrowed as a whole directly from the

classical languages Greek and Latin. Instead, they have been coined in

the course of time by combining these root morphemes which the language

user can discover through comparison of already existing complex

words. In fact, neo-classical compounding has given rise to a huge

86 word-formation

pan-European lexicon. European languages share huge parts of the set of

neo-classical compounds which are used in science, government, culture,

and business.

The example tele-camera in (15b) shows that an initial combining form

can also combine with words. We are not always sure if the second

constituent is a lexeme or a combining form that corresponds in form to

a lexeme. In the case of television, for instance, there is a lexeme vision, but

this lexeme has a more speciWc meaning than just ‘‘sight’’ (the meaning of

-vision as a combining form). The same applies to graph, as pointed out in

the preceding paragraph. This is why English dictionaries may have two

entries, one for -graph as a combining form, and one for graph as a word.

The words in (15c) show that there are also cases in which a lexeme is

followed by a Wnal combining form. In bureaucrat, the Wrst constituent

bureau is a lexeme, unlike -crat. The word magneto-metry is another

instantiation of this pattern since there is a lexeme magnet. Note, however,

that the form of the lexeme magnet is special in that it is followed by the

linking vowel o. Indeed, most words when used in the Wrst position of

a neo-classical compound have a linking vowel. The word magneto-

hydrodynamic shows that we also Wnd neo-classical compounds with more

than two constituents.

Final combining forms such as -logy and -graphy are perhaps not to be

considered as one, but as two morphemes since they lend themselves to

further morphological analysis, and may be divided into log-y and graph-y

respectively, as suggested by a comparison with words such as bio-log-ist

and geo-graph-er.

When a combining form is combined with a lexeme, the lexeme is usually

taken from the non-native stratum of the lexicon. This is particularly the

case for Wnal combining forms. For instance, the Dutch neo-classical

compound hond-o-loog ‘‘dog specialist’’ derived from the native word

hond ‘‘dog’’ is felt as a jocular type of word-formation. However, initial

combining forms such as the Greek roots eco- and tele- are often attached

to native words as well, as illustrated here for Dutch:

(16) eco-ontbijt ‘‘eco-breakfast’’

eco-paddestoel ‘‘eco-mushroom’’

eco-sigaret ‘‘eco-cigarette’’

tele-leren ‘‘tele-learning’’

tele-werken ‘‘tele-working’’

tele-winkelen ‘‘tele-shopping’’

compounding 87

One might conclude from such data that these initial combining forms have

become preWxes.

A lot of initial combining forms have been created by means of trunca-

tion. A good example is the pan-European initial combining form

euro- that is a truncation of the word Europa. Other examples of such

truncations are afro-, compu-, crea-, cine-, cyber-, digi-, docu-, and Xexi- that

are all used very productively to coin new words in most European

languages.

In short, neo-classical compounding is a very important source of new

words, even though it does not combine lexemes, and therefore does not

possess the same degree of semantic transparency as regular compounds.

The observations on neo-classical compounding in the preceding para-

graphs all lead to the conclusion that it is not always easy or possible to

decide if a complex word is a case of compounding or of derivation.

4.4 Interfixes and allomorphy

When lexemes are used as building blocks of compounds, they may exhibit

a special form. This is the case for Greek compounds: in most cases the Wrst

constituent ends in the vowel /o/ which is added to the stem-form of the

lexeme. These vowels are clearly diVerent from the inXectional endings that

appear in the corresponding words (Ralli 1992: 145, 153):

(17) pag-o-vuno ‘‘ice mountain, ice berg’’ < pag-os ‘‘ice’’, vun-o ‘‘mountain’’

psom-o-tiri ‘‘bread (and) cheese’’ < psom-i ‘‘bread’’, tir-i ‘‘cheese’’

sime-o-stolizmos ‘‘Xag decoration’’ < sime-a ‘‘Xag’’, stolizm-os ‘‘decoration’’

Such vowels, which are called interWxes or linking elements, do not con-

tribute a meaning of their own to the word as a whole, but only function to

create stem-forms that are suitable for being used in compounds. Phono-

logically, they belong to the Wrst constituent of the compound. This can be

concluded from the division of such words into syllables: when possible,

these vowels form a syllable together with the last consonant of the Wrst

stem, for instance pa.go.vu.no (the dots indicate syllable boundaries). If the

linking element did not belong phonologically to the Wrst stem, we would

have expected the syllabiWcation pattern pag.o.vu.no. The appearance of

these linking elements can be interpreted as a case of stem allomorphy,

the phenomenon that stems may have more than one phonological form.

88 word-formation

The selection of a particular allomorph is governed by the morphological

context. In the case ofGreek compounds, we thus have to state that the stem

allomorph that ends in -o has to be used as Wrst constituent of such words.

Stem allomorphy is also found in Dutch compounds. Consider the

following cases (Booij 2002a: 178–80):

(18) schaap ‘‘sheep’’ schaap-herder ‘‘shepherd’’

schaap-s-kop ‘‘sheep’s head’’

schap-en-vlees ‘‘lit. sheep’s meat, mutton’’

kind ‘‘child’’ kind-er-wagen ‘‘lit. children’s cart, pram’’

koningin ‘‘queen’’ koninginn-e-dag ‘‘Queen’s birthday’’

These examples illustrate the use of the linking elements -s, -

e

(mostly spelt

as -e or -en), and -er. Historically, these are mainly inXectional endings. For

instance, -s derives from a genitive suYx -s, and -er is an old plural suYx

that is still used as such in German. The -e is historically either a case suYx

or the last vowel of the stem. Synchronically, these linking elements

no longer have the status of inXectional suYxes. Since they belong phono-

logically to the Wrst stem, we may say that, for instance, the lexeme schaap

has three stem allomorphs: schaap-, schaaps-, and schape-. When a native

speaker of Dutch coins new compounds, the choice of a particular

allomorph will often be based on analogy. Since we have koninginnedag,

we will also opt for the schwa-Wnal allomorph in a new compound such as

koninginne-hoed ‘‘queen’s hat’’. In the case of schaap, however, all three

allomorphs may be chosen. The morphological structure of the left

constituent may also play a role. For instance, Dutch diminutive nouns,

which end in -je, are always followed by the linking element -s when used as

the Wrst constituent of a compound:

(19) meisje-s-lijk ‘‘girl’s corpse’’, *meisje-lijk

dagje-s-mens ‘‘day tripper’’, *dagje-mens

rijtje-s-huis ‘‘row house’’, *rijtje-huis

Dutch has a plural suYx -s as well. However, the -s in these compounds

cannot be interpreted as a plural suYx, given their meaning. For instance,

meisjeslijk denotes the corpse of only one girl, and dagjesmens denotes a

person who takes a one-day trip.

The choice of stem allomorph in Dutch compounds is also constrained

by a paradigmatic factor: one can only choose the schwa-Wnal allomorph

(spelt with Wnal -en) if the plural suYx for the relevant noun is the suYx

compounding 89

-en. The appearance of the linking element -s is not subject to such a

constraint. The following data illustrate this regularity:

(20) Noun Plural form Compound

varken ‘‘pig’’ varken-s varken-s-vlees ‘‘pig’s meat’’, *varken-e-vlees

leraar ‘‘teacher’’ lerar-en leraar-s-salaris ‘‘teacher’s salary’’,

lerar-en-vergadering ‘‘teachers’ meeting’’

The use of linking elements is not restricted to nominal compounds.

In Spanish the vowel i can be used as an interWx in coordinative adjectival

compounds (Rainer and Varela 1992: 132):

(21) roj-i-blanco ‘‘red and white’’ < rojo ‘‘red’’, blanco ‘‘white’’

clar-i-vidente ‘‘clairvoyant’’ < clar-o ‘‘clear’’, vidente ‘‘seeing’’

This kind of stem allomorphy is thus a variation in the shape of morphemes

that is not governed by the phonology of a language, but is regulated by its

morphology.

4.5 Synthetic compounds and noun incorporation

Consider the following English nominal compounds of which the head is

a deverbal noun:

(22) sword-swallower, heart-breaker, church-goer, money-changer, typesetter

These compounds pose some analytical questions. First, some of the nominal

heads such as swallower and goer do not occur as words of their own. These

are possible, but not established English words. Thus, these words show that

possible words can function as building blocks inword-formation.Onemight

also argue that thesewords are derived by attaching the suYx -er to the verbal

compounds sword-swallow, heart-break, etc. This alternative analysis is inad-

equate because verbal compounding is not a productive process in English,

and hence does not license the possible words sword-swallow or heart-break.

What we see here is that the use of one word-formation process, nominal

compounding, implies the use of another word-formation process, deverbal

nominalization with -er, which provides possible words like swallower and

breaker. These words are then used as the heads of nominal compounds. The

term synthetic compounding is traditionally used to indicate that this kind

of word-formation looks like the simultaneous use of compounding and

90 word-formation

derivation. This instance of simultaneous use can be expressed through the

conXation of the English morphological templates for NN compounds and

deverbal nouns into the following one:

(23) [[N] [V-er]N]N

Such conXation of compounding and derivation is also found for other

types of word-formation, as illustrated by blue-eyed, in which AA

compounding is conXated with the formation of denominal adjectives

(note that eyed itself is not an established word of English).

The second special property of the compounds in (22) is that the left

constituent fulWls a speciWc semantic role with respect to the verbal base of

the right constituent. For instance, in sword swallower, the constituent

sword functions as the Patient-argument of swallow, the verbal base of

the head noun. In order to understand this semantic regularity, we must

look in more detail at how semantic roles are linked to verbs. Verbs assign

speciWc semantic roles to the NPs that they occur with in a clause. The verb

to swallow may occur with two NPs, one with the role of Agent (the

controller of the action), and one with the role of Patient (the entity that

undergoes the action mentioned by the verb). NPs that receive a semantic

role from a verb are called the arguments of that verb. Similarly, in the

synthetic compound church goer, the constituent church fulWls the semantic

role of Goal with respect to the verbal base go of the head noun goer

because to go assigns the semantic role of Goal. The question thus arises

how we account for this semantic relationship between the left noun and

the verbal base of the right noun.

Is it possible to consider these words to be cases of -er-aYxation to VPs

such as to swallow swords? If this were possible, we would have explained

why the left nouns in these compounds have the same semantic roles as

those they receive in verb phrases. The answer is negative. This guess may

look adequate, but from a formal point of view it is a hopeless idea. First, in

the compound, the word sword appears in its bare form, as a stem, which is

impossible in a phrase: *to swallow sword. Second, the word order would

be wrong. In English VPs the syntactic object follows the verb, but in sword

swallower the opposite order is found. Therefore, it is a better option to

analyse these words as regular compounds, with the special property that

the argument structure of the verbal base is inherited by the derived noun

with the suYx -er. Thus, the head noun can assign a semantic role such as

Patient or Goal to the left constituent.

compounding 91

As mentioned before, most Germanic languages do not have productive

processes for verbal compounding. However, English gerund forms and

Dutch inWnitival forms of verbs do occur as the heads of compounds:

(24) English: mountain-climbing, word-processing

Dutch: hard-lopen ‘‘fast walking’’, wedstrijd-zwemmen ‘‘competition

swimming’’

These words, however, do not have Wnite forms, and the sentence I went

mountain-climbing is much better than I mountain-climb every Saturday

(Mithun 1984: 847). Speakers of Dutch use the periphrastic progressive

construction to solve the problem caused by the lack of Wnite forms, as in Ik

ben aan het hardlopen ‘‘I am fast-walking’’ (compare the ungrammatical

sentence*Ikhard-loop ‘‘I fast-walk’’). InanotherGermanic language,Frisian,

occasionally newNV compounds with Wnite forms do occur (Dyk 1997: 29):

(25) Hja bole-bakt al jieren met njocht

she loaf-bakes already years with pleasure

‘‘She bakes loaves already for many years with pleasure’’

The combination of a noun and a verb into a verbal compound is usually

called noun incorporation. It is used primarily to form verbs that express an

institutionalized activity. These incorporated nouns do not denote speciWc

objects, they are non-referential. In example (25), the noun bole does not

refer to a speciWc loaf, but to loaves in general. Hence such incorporated

nouns are non-referential.

The diVerence between a verb phrase with an NP and noun incorpor-

ation is illustrated by the following example from the Micronesian lan-

guage Ponapean (Mithun 1984: 850):

(26) a. I kanga-la wini-o

I eat-comp medicine-that

‘‘I took all that medicine’’

b. I keng-winih-la

I eat-medicine-comp

‘‘I completed my medicine-taking’’

In (26b) we see a case of noun incorporation. It has a diVerent meaning than

(26a), which has a syntactically independent object wini-o. In (26b), with an

incorporated and non-referential object winih-, the sentence indicates com-

pletion of the action of medicine taking, while there may be medicine left.

92 word-formation

Typically, the nouns in these cases of incorporation are unmarked for

deWniteness, number or case, and the verbal compound behaves as an

intransitive verb, whereas its verbal head is transitive. Thus, noun incor-

poration often has the eVect of creating verbs with reduced syntactic valency:

since the Patient-argument of the verb is expressed by the incorporated noun,

this argument will no longer receive an independent syntactic expression.

Summary

Compoundsarecombinationsof twoormore lexemes. Inmany languages this

word-formation process is used frequently because of its semantic transpare-

ncy and versatility. The boundary between compounds and phrasal lexi-

cal expressions is not always clear. Diagnostics for the phrasal status of lexical

expressions are internal inXection, and the splittability of their constituents.

The distinction between compounding and derivation is sometimes

blurred because lexemes as parts of compounds may receive specialized

interpretations. Such compound constituents are aYx-like, since aYxes

also depend on occurring as part of complex words for them to receive

a proper semantic interpretation. The use of Greek and Latin roots for

neo-classical word-formation also relativizes the distinction between

compounding and word-derivation, because classical roots that are not

lexemes might be considered aYxes.

Stem allomorphy is a recurrent phenomenon in compounding. The

choice of a particular linking element may be determined by paradigmatic

factors such as analogy to existing compounds and the shape of the plural

form of the corresponding lexeme.

Synthetic compounding can be seen as the simultaneous use of com-

pounding and derivational morphology in coining a new compound. The

semantic role of the non-head in these compounds may be determined by

the argument structure of the verbal base of the head noun. Noun incorp-

oration is a similar process.

Questions

1. Which categories of words can combine in English compounds? Whatrestrictions on possible combinations of categories do you observe?

compounding 93

2. Give the complete morphological structure of the following English com-pounds: recreation hall, book keeping, truck driver, pickpocket, underdog,homegrown.

3. Dutch appears to make use of noun incorporation, as illustrated by thefollowing verbal compounds (Weggelaar 1986):

klapper-tand ‘‘to have chattering teeth’’ < klapper ‘‘to rattle’’, tand ‘‘tooth’’stamp-voet ‘‘to stamp one’s feet’’ < stamp ‘‘to stamp’’, < voet ‘‘feet’’.

What problem do these compounds pose for the Right-hand Head Rulewhich is normally applicable to Dutch compounds?

4. Consider the following recently coined words ending in gate coined onanalogy to Watergate: Irangate, Monicagate, nipplegate. These three wordsdenote scandals. Are these words cases of compounding, or of derivation?Explain your answer.

5. Compare the following two sentences from Yucatec, a Mexican language(Mithun 1984: 858):

a. k-in-c’ak-Ø-k ce’ icil in-koolINCOMP-I-chop-it-IMPF tree in my-cornfield‘‘I chop the tree in my cornfield’’

b. k-in-c’ak-ce’-t-ik in-koolINCOMP-I-chop-tree-TR-IMPF my-cornfield‘‘I clear my cornfield’’

Explain the differences in interpretation between these two almost identicalsentences.

6. Compare the following two adjective–noun combinations in ModernGreek:

[aghri-o]A [ghat-os]N ‘‘wild cat’’ < aghri- ‘‘wild’’ + ghatos ‘‘cat’’[psixr-os]A [polem-os]N ‘‘cold war’’ < psixros ‘‘cold’’ + polemos ‘‘war’’

In the first example, there is a linking element o, and only an inflectionalending on the second word; in the second AN combination, both constitu-ents are inflected. How can this difference in inflectional behaviour beexplained?

7. Consider the following Italian verb–noun compound: [[spazza]V[camino]N]N‘‘lit. sweep chimney, chimney sweep(er)’’. Is this an endocentric or anexocentric compound?

8. Try to specify the semantic relation between the two constituents of therecently coined English compounds bear jam, deprivation cuisine, flash mob,information pollution, man breasts, office creeper, salad dodger. Which ofthem cannot be interpreted easily on the basis of the meanings of theirconstituent words? (source www.wordspy.com)

9. In Punjabi, there are noun–noun compounds that denote the superordinatesemantic category of the two nouns involved (Bhatia 1993: 320):

94 word-formation

www.wordspy.com

hatth-pair ‘‘body’’ < hatth ‘‘hand’’, pair ‘‘feet’’uu-nakk ‘‘face’’ < muu ‘‘mouth’’, nakk ‘‘nose’’bas-kaar ‘‘vehicle’’ < bas ‘‘bus’’, kaar ‘‘car’’

What problem do these compounds pose for the hypothesis that com-pounds always have one of their constituents as a head?

10. Which formal processes are involved in the formation of the followingcomplex words in Punjabi (Bhatia 1993: 322)?

paanii ‘‘water’’ paanii-shaanii ‘‘water and the like’’kamm ‘‘work’’ kamm-shamm ‘‘work and the like’’gapp ‘‘silly talk’’ gapp-shapp ‘‘silly talk and the rest’’ultaa ‘‘conflicting nonsense’’ ultaa-shultaa ‘‘conflicting and other

nonsense’’

Further reading

For a survey of types of compounding, see Olsen (2000, 2001); a survey ofcompounding in a number of European languages is given in Scalise (1992b).A number of articles on separable complex verbs can be found in YoM 2003.The problem of stem selection is discussed in Aronoff (1994). The role ofanalogy in selecting a linking element for Dutch compounds is shown inKrott (2001).There is a wealth of literature on synthetic compounds. Botha (1984) pro-

vides a survey of the debate. Specific contributions to the debate are Selkirk(1982), Hoeksema (1985), Booij (1988), and Hoekstra and van der Putten(1988). For different views of noun incorporation, see Mithun (1984, 1999,2000) and Rosen (1989) for a lexical/morphological, and Baker (1988, 1996,2001) for a syntactic analysis.

compounding 95


Part IIIInflection


5Inflection

5.1 Inflectional properties

InXection is themorphological marking of properties on a lexeme resulting in

a number of forms for that lexeme, a set of grammatical words. In the

example given in section 2.1, the inXection of the Polish lexeme kot, each

form of the lexeme has properties with respect to two inXectional dimensions,

number and case. These two inXectional dimensions for nouns are found in

many languages. The dimensions are referred to as morphosyntactic categor-

ies because they may play a role both in morphology and in syntax. For each

dimension or category, there is more than one value. In the case of Polish

nouns, there are two values for number: singular and plural, and seven

diVerent values for case: nominative, genitive, dative, accusative, instru-

mental, locative, and vocative. These values are referred to asmorphosyn-

tactic features. A particular cell in the paradigm of kot is thus Wlled with a

word form with a speciWc set of morphosyntactic features. The nominative

plural form koty ‘‘cats’’ can be represented as follows:

(1) kot-y number: plural

case: nominative

Another category of words that plays a prominent role in inXection are

verbs, which are often inXected for categories such as tense, aspect, and

5.1 InXectional properties 99

5.2 The roles of inXection 103

5.3 InXection and derivation 112

5.4 Theoretical models 115

5.5 Morpheme order 119

Summary 122

Questions 122

Further reading 124

mood. Some of us have been introduced to this kind of inXectional morph-

ology through textbooks on Latin that present the whole set of verbal

forms in a number of paradigms. A part of the paradigm of laudare ‘‘to

praise’’ as presented in such textbooks is given in Table 5.1 (the macron

over the vowel letter indicates length). The three labels present, imperfect,

and perfect are the traditional labels for these Latin verbal forms (cf.

section 6.2). Each of these forms expresses a property for the categories

tense, aspect, and mood. For instance, the form laudat ‘‘he praises’’ has

the following values for these three verbal categories (in addition, it has

values for the number and person of the subject of this verb):

(2) laud-at tense: present

aspect: imperfective

mood: indicative

number: singular

person: 3

The following list presents a (non-exhaustive) survey of the diVerent

types of morphosyntactic information that are found as morphological

markings on nouns, verbs, and adjectives in the languages of the world:

(3) Nouns: Number (singular, plural, dual, etc.), Case (nominative, genitive,

accusative, etc.), DeWniteness, Gender;

Verbs: Tense (present, past, future), Aspect (imperfective, perfective, etc.),

Mood (indicative, subjunctive, imperative, etc.), Voice (active, passive, etc.),

Number (singular, plural, etc.), Person (Wrst, second, third), Gender;

Adjectives: Degree (positive, comparative, superlative), Number, Gender,

Case, DeWniteness.

The category Voice for verbs comprises distinctions such as that between

active and passive voice. The Latin word laudat belongs to the active voice,

Table 5.1. Indicative forms of the Latin verb laudare

indicative present indicative imperfect indicative perfect

1sg laudo laudabam laudavi

2sg laudas laudabas laudavisti

3sg laudat laudabat laudavit

1pl laudamus laudabamus laudavimus

2pl laudatis laudabatis laudavistis

3pl laudant laudabant laudaverunt

100 inflection

but Latin also has special verbal forms with a passive interpretation such as

laudatur ‘‘he is praised’’.

The category Degree for adjectives comprises the positive, comparative,

and superlative degree, as in English happy–happier–happiest. Some

languages also have an equative form for adjectives, as is the case for

Finnish (Sulkala and Karjalainen 1992: 172):

(4) Jukka on Peka-n pit-uinen

Jukka be.3sg Pekka-gen long-equative

‘‘Jukka is of the same length as Pekka’’

In addition, the form of an adjective may be determined by a relation of

agreement with its head noun and determiner with respect to properties

such as deWniteness, gender, number, and case.

An additional distinction in the domain of verbal inXection is that

between Wnite forms, which are inXected for the categories tense, number,

and person, versus inWnite (or non-Wnite) forms such as inWnitives, gerunds,

and participles. The following Dutch sentences illustrate the use of inWnitive

and participle forms:

(5) a. Jan wil kom-en

John want.pres.3sg come-inf

‘‘John wants to come’’

b. Huil-end vertrok hij

cry-pres.ptcp leave.past.3sg he

‘‘He left crying’’

As you can see, these non-Wnite verbs are used in sentences that have a Wnite

verbal form as well. Their actual use is a matter of complex syntactic

principles, and will not be discussed here.

A remarkable property of these non-Wnite forms is that they do not only

have verbal properties, but also properties of other lexical categories such

as nouns and adjectives. For instance, inWnitives can be described as

nominal forms of verbs. In some languages, they may be preceded by a

determiner or a preposition, and they may function as the head of NPs, as

illustrated here for the Dutch inWnitive:

(6) a. Jan houdt van het lez-en van poezie

John likes of the read-inf of poetry

‘‘John likes reading poetry’’

inflection 101

b. Jan houdt van lez-en

John likes of read-inf

‘‘John likes reading’’

The English word reading in the gloss of sentence (6a) is an example of a

gerund form. As shown by the phrase John’s reading poetry ‘‘the reading of

poetry by John’’, gerunds, like inWnitives, have nominal properties: the

preceding noun John has the same form John’s that we Wnd in the phrase

John’s book. At the same time, this gerund form still has verbal properties: it

combines with the preposition-less complement poetry. The occurrence with

preposition-less complements is a characteristic feature of verbs. When one

adds a complement to a noun, that complement has to be preceded by a

preposition, as in John’s love of poetry where the preposition of has to

precede the complement noun poetry (*John’s love poetry is illformed).

Participles are verbal forms that can be used as adjectives, and are also

inXected as such in relevant syntactic contexts. In the German example in

(7) below, the present participle has the adjectival ending -e that is required

for attributive adjectives in deWnite NPs with a masculine head noun

(Haspelmath 1996: 44):

(7) Der im Wald laut sing-end-e Wanderer

the in.the forest loud sing-pres.ptcp-e hiker

‘‘the hiker who is singing loud in the forest’’

A less known non-Wnite verbal form with adverbial properties is the

converb. Converbs are verbal forms used as adverbs, as in the following

sentence from Kannada, a Dravidian language (Haspelmath 1996: 50):

(8) Yaar-ig-uu heel-ade eke bande

who-dat-indef say-neg.cvb why come.pret.2sg

‘‘Why did you come without telling anyone?’’

What is remarkable about these non-Wnite forms is that their inXection

appears to be category-changing, from verb to noun, adjective, or adverb,

without erasing the verbal properties of these forms. This is remarkable

because, typically, inXection does not aVect the syntactic category of a

word, unlike derivation.

In many languages, nouns and verbs are classiWed into a number of

inXectional classes, called declensions in the case of nouns and adjectives,

and conjugations in the case of verbs. These declensions and conjugations do

not express morphosyntactic properties themselves, but determine how

102 inflection

such properties are expressed. Latin, for instance, has Wve declension

classes for nouns that determine number and case marking, as illustrated

in (9) for the singular forms in the nominative and genitive case:

(9) Class nom.sg gen.sg Words with the same inXectional pattern

I mensa ‘‘table’’ mensae puella ‘‘girl’’, colonia ‘‘colony’’

II hortus ‘‘garden’’ horti captivus ‘‘prisoner’’, buxus ‘‘buxus tree’’

III rex ‘‘king’’ regis onus ‘‘burden’’, civis ‘‘citizen’’

IV fructus ‘‘fruit’’ fructus domus ‘‘house’’, manus ‘‘hand’’

V dies ‘‘day’’ diei res ‘‘thing’’, meridies ‘‘afternoon’’

Latin adjectives are also divided into declension classes, but there are only

three declensions for adjectives compared to Wve for nouns.

Latin is also a languagewith inXectional classes in the verbal domain. It has

four conjugations for its verbs, with diVerent thematic vowels after the verbal

root. The forms listed in Table 5.1 above all have a thematic vowel a (except

laudo, where the thematic vowel is absent before the ending o). Thus, laudare

belongs to the conjugation of a-verbs. Examples of verbs of the three other

conjugations are delere ‘‘to delete’’, emere ‘‘to buy’’, and audıre ‘‘to hear’’.

5.2 The roles of inflection

Consider the following German sentence that illustrates the roles of inXec-

tion in constructing a sentence of German:

(10) Der Vater putz-t sein-er

the.masc.sg.nom father.masc.sg.nom brush-pres.3sg his-fem.sg.dat

Tochter die Schuh-e

daughter.fem.sg.dat the.pl.acc shoe.masc-pl.acc

‘‘The father brushes the shoes for his daughter’’

The words Vater, Tochter, and Schuhe are marked for number. The Wrst

two nouns are marked as singular, the last one as plural. These three nouns

are also marked for case. Since der Vater ‘‘the father’’ is the subject of the

sentence, it must be marked with nominative case. The noun Tochter

‘‘daughter’’ has to be marked with dative case, because seiner Tochter

functions as indirect object of the sentence. Finally, the noun Schuh-e has

the accusative form since die Schuhe functions as the direct object of this

inflection 103

sentence. The case marking of these nouns is obviously determined by the

syntactic contexts in which they occur, and hence we call this contextual

inXection. It stands in contrast with the number marking for these nouns,

which is not determined by the syntactic context. The choice of a particular

number is determined by what information the speaker wants to convey,

and we therefore call it inherent inXection. The present tense on the verb is

also a case of inherent inXection: it is a matter of free choice: the verb could

as well have appeared with a diVerent tense. Below, we will see that the

distinction between inherent and contextual inXection plays an important

role in analysing morphological systems.

The basic role of contextual inXection is to mark the relationship

between a head and a dependent in a syntactic construction. Two basic

types of dependency must be distinguished: government and agreement. Let

us Wrst focus on government.

In languages with case systems, the verb assigns a certain case to each of

its dependent NPs such as the subject, the object, and the indirect object.

This is illustrated by the German sentence (10), and also by the following

Latin sentence, (25) from Chapter 2:

(11) Manu-s manu-m lava-t

hand-nom.sg hand-acc.sg wash-3sg

‘‘One hand washes the other’’

It is the verb that is the head of the clause since the verb determines that

there must be a subject and, in this case, an object. The case markings in

sentences (10) and (11) are instances of dependent marking since the syn-

tactic relations between the verb and the nouns are marked on the depend-

ents. This kind of dependent marking is a case of government. We speak of

government when a constituent imposes requirements on a related con-

stituent. In the example (11), the verb requires speciWc case markings on its

subject and object. The verb itself, however, bears no corresponding case

markings, and that is why we qualify this kind of case marking as govern-

ment. In contrast, in the case of agreement both constituents involved

are marked for the relevant inXectional properties. For instance, in the

phrase seiner Tochter in (10) the head noun Tochter and the dependent

modiWer seiner have the same dative case, and hence the inXection of

seiner is an instance of agreement rather than government.

The speciWc case form of the head noun of a noun phrase not only marks

its dependency on the verb, but also indicates a speciWc semantic relation

104 inflection

of the noun phrase to the verb. In (11), the NP manus denotes the Agent of

the verb lavare ‘‘to wash’’, whereasmanum ‘‘hand’’ denotes the Patient that

undergoes the action. When the verb governs an NP with the role of

recipient, this NP will be marked by means of the dative case, as in (10).

These cases are called structural casesor direct cases. They stand inopposition

to inherent or semantic cases. Semantic case is illustrated by the following

Latin sentence in which the ablative case is used for creating the adverbial

instrumental phrase ‘‘with a knife’’, whereas the accusative case is a

structural case, and marks the direct object ‘‘the bread’’:

(12) Cultr-o pan-em secat

knife-sg.abl bread-sg.acc cuts

‘‘He cuts the bread with a knife’’

The non-nominative structural cases are sometimes called the oblique cases.

These are the cases that are required by a speciWc syntactic structure (accusa-

tive and dative case), whereas the nominative is the case form thatmay also be

used without a speciWc syntactic context, for instance as a citation form.

The assignment of case may be lexically governed, and is called lexical

case. Accusative case is the default case that an NP with the syntactic

function of direct object will receive. Verbs may, however, require another

case on their object-NP. For instance, in eighteenth-century German, the

verb kennen ‘‘to know’’ marked its object with genitive case. A famous

sentence from Bach’s Matthew’s Passion sung by Petrus is (13a); its form

in present-day German is (13b):

(13) a. Ich kenne de-s Mensch-en nicht

I know the-masc.gen.sg man-masc.gen.sg not

‘‘I do not know the man’’

b. Ich kenne de-n Mensch-en nicht

I know the-masc.acc.sg man-masc.acc.sg not

‘‘I do not know the man’’

In (13a) the object is marked as genitive. In present-day German this lexical

marking with genitive case for the object of the verb kennen ‘‘to know’’ has

disappeared, and hence the default accusative case is assigned. The German

verb trauen ‘‘to trust’’ requires dative case on its object, as inWir trauen ihm

‘‘We trust him’’, with the dative form ihm of the 3sg personal pronoun.

Case forms that are used in structural case assignmentmay also function as

semantic cases. For instance, Romam, the accusative form of the Latin noun

inflection 105

Roma, can be used as a directional phrase with themeaning ‘‘to Rome’’, as in

Eo Romam ‘‘I go to Rome’’. Both in Latin andGerman, the accusuative case

is also used to mark durational phrases, as in German:

(14) Wir laufen den ganz-en Tag

we walk the.masc.acc.sg whole-masc.acc.sg day.masc.acc.sg

‘‘We walk the whole day’’

The durational phrase den ganzen Tag has the accusative form.

Another well-known case of dependent marking in Indo-European lan-

guages is that prepositions govern the choice of case form of the noun with

which they combine. German prepositions can be classiWed according to

the kind of case marking they require on their dependent nouns. For

instance, the following German prepositions always require accusative

case: fur ‘‘for’’, durch ‘‘through’’, bis ‘‘until’’, gegen ‘‘against’’, ohne ‘‘with-

out’’, and um ‘‘around’’. Other German prepositions appear with either

dative or accusative case, depending on the meaning expressed:

(15) a. Ich stecke die Zeitung hinter den Spiegel

I put the newspaper behind the.acc.sg mirror

‘‘I put the newspaper behind the mirror’’

b. Die Zeitung steckt hinter dem Spiegel

the newspaper is behind the.dat.sg mirror

‘‘The newspaper is behind the mirror’’

Although prepositions govern their NP-complement with respect to case,

in some cases there is a choice, which depends on the particular semantic

relation between the verb and the preposition. If the verb stecken ‘‘to put’’

denotes an action (15a), the preposition hinter requires accusative case, if it

expresses a state (15b), the same preposition hinter requires dative case.

Nouns may require a particular marking on their dependent. In the

phrase John’s house the dependent NP John is marked as such through

the presence of the possessive marker s, whereas the head noun house is not

marked for this syntactic relationship. In languages with genitive case, this

case is typically used for marking relations between nouns.

In some languages it is not the dependent, but the head that is morpho-

logically marked for standing in a syntactic relationship to its dependent.

For instance, in Hungarian, marking is on the head (Nichols 1986: 57):

(16) az ember haz-a

the man house-3sg

‘‘the man’s house’’

106 inflection

Semitic languages are well known for their head marking patterns: the head

noun of an NP has often a special phonological form when it is followed

by a dependent NP. The word bayit ‘‘house’’ in Modern Hebrew has the

form beyt when followed by an NP-complement like sefer ‘‘book’’

(Borer 1988: 50):

(17) beyt sefer

house book

‘‘school’’

This special form of the head noun is called the construct state since it

appears when the noun is in construction with another noun. In this case,

the head is marked by means of allomorphy, instead of by an additional

morpheme.

The second main type of contextual inXection besides government

is agreement (also called concord). In many Indo-European languages

attributive adjectives agree with respect to a number of morphosyntactic

properties with their head nouns, as is illustrated by the following examples

(Latin and Hebrew are adapted from Barlow and Ferguson 1988: 3, 5). The

nouns function as the source of the shared property, the adjectives and

determiners as the targets.

(18) a. Latin: acta vir-orum omni-um bon-orum

deeds man.masc-pl.gen all-pl.gen good-masc.pl.gen

‘‘deeds of all good men’’

b. Hebrew: ha-isha ha-tov-a

the-woman. fem.sg the-good-fem.sg

‘‘the good woman’’

c. Dutch: het oud-e paard

the.def.sg.neut old-def.sg.neut horse.sg.neut

‘‘the old horse’’

d. French: les femme-s enchante-e-s

the women.fem-pl delighted-fem-pl

‘‘the delighted women’’

In these examples it is the dependent adjective that is marked for properties

of the head noun with respect to case, number, and gender. In the Hebrew

and the Dutch case, the feature definite of the noun phrase assigned by the

determiners ha and het respectively, is also expressed on the adjective.

Example (18a) shows that in agreement contexts both inherently and con-

textually determined morphosyntactic properties can be expressed on the

inflection 107

constituent that has to be marked. In addition to the inherent inXectional

features masculine and plural, the contextually determined genitive

case marking of the head noun vir ‘‘man’’ is also expressed on the following

quantiWer and adjective.

In some languages, such as those of the Indo-European family, there is

no direct morphological marking of gender on nouns. For such languages,

we only observe direct morphological eVects of gender on the form of

dependent adjectives and determiners. In French, the gender of a noun

(masculine or feminine) manifests itself in the choice of the sg

articles (indeWnite un or une, deWnite le or la), and in the form of the

adjective (suYx -e in feminine forms), but not in the inXectional form of

the noun itself. At Wrst sight, it may seem as if nouns in languages such as

Latin do express gender in their inXectional form. For instance, most Latin

nouns ending in -a are feminine. Yet, the ending -a is no direct marking of

gender, but is characteristic for a speciWc declension class of Latin nouns,

and there is a correlation between declension class and gender: most nouns

of the -a declension class are feminine. Words such as nauta ‘‘sailor’’ and

poeta ‘‘poet’’, however, are masculine notwithstanding the presence of the

ending -a, as we can tell from gender agreement. It is nauta bonus ‘‘the good

sailor’’ where bonus is the masc.sg.nom form, not *nauta bona.

When gender of nouns manifests itself indirectly, it is perhaps more

appropriate to consider the marking of gender on adjectives as a case of

government (instead of agreement), parallel to the government of speciWc

case forms of nouns by adpositions. For instance, we do not say that the

German preposition durch itself has accusative case, and that the noun

governed by it agrees with respect to case. Rather, we say that durch

requires a speciWc case form of the noun. Similarly, we might say that a

masculine noun selects a masculine form of its adjective and determiner.

Dependent marking is also found on relative pronouns that function as

the subject of a relative clause. In Dutch, for instance, these pronouns have

to agree in number and gender with their antecedents:

(19) a. de jongen die ziek is

the boy.masc.sg who ill is

‘‘the boy who is ill’’

b. het meisje dat ziek is

the.neut.sg.def girl.neut.sg who.neut.sg ill is

‘‘the girl who is ill’’

108 inflection

The interlinearmorphemicglossing in the examples (19) shows thatagreement

is not amatter of identity of features, but of non-contradictoriness of features.

The relative pronoun die can beused for non-neuter singular antecedents, and

for all plural antecedents. It is only the relative pronoun dat that has a speciWc

feature set, neutral and singular. The pronoun die is used in all other

cases. It is the default form, and hence it is unspeciWed for the categories

gender and number. The same applies to the article het which is only used

in definite sg.neutNPs, whereas de is the default article for deWnite NPs.

These examples also illustrate that agreement is not always marked by

means of morphology, but may also be marked through the choice of a

speciWc lexical item (in this example die or dat). This was also the case for

the Hebrew construct state in (17), a case of government. This is in

particular true for pronouns, which often have diVerent lexemes for diVer-

ent morphosyntactic properties. The examples in (19) also illustrate that

‘gender’ is primarily a formal, and not a semantic category: the wordmeisje

‘‘girl’’ denotes a person of the female sex, yet it is a neuter noun that selects

the neuter relative pronoun.

As illustrated by the Dutch sentence (18c), the feature (in)definite may

play a role in agreement between an adjective and its head noun. In

German, attributive adjectives have two patterns of contextual inXection,

traditionally called weak inXection and strong inXection. In weak inXection

there are less formal diVerences between the cells of a paradigm than in

strong inXection. When an adjective is not preceded by an article, it is

subject to strong inXection. (When it is preceded by an indeWnite article it

has a slightly simpliWed form of strong inXection, called mixed inXection.)

After a deWnite article it has weak inXection. This is illustrated here for NPs

with nominative case, where there is no diVerence between strong and

mixed inXection (Eisenberg 1994: 235):

(20) strong heiss-er Tee hot-masc.nom.sg tea.masc.nom.sg ‘‘hot tea’’

heiss-e Suppe hot-fem.nom.sg soup.fem.nom.sg ‘‘hot soup’’

heiss-es Wasser hot-neut.nom.sgwater.neut.nom.sg ‘‘hot water’’

weak der heiss-e Tee the hot-masc.nom.sg tea.masc.nom.sg ‘‘the hot

tea’’

die heiss-e Suppe the hot-fem.nom.sg soup.fem.nom.sg ‘‘the hot

soup’’

das heiss-e Wasser the hot-neut.nom.sg water.neut.nom.sg ‘‘the

hot water’’

inflection 109

As the examples show, there are more formal diVerences in the adjectival

forms of strong inXection than in those of weak inXection.

Agreement is not always a case of dependent marking, as we saw above.

A frequent type of agreement in Indo-European languages is subject–verb

agreement. The relevantmorphosyntactic properties are those of number and

person. This is a case of head marking since the verb is the head, and the

subject is the dependent. Subject–verb agreement is therefore an exception to

the tendency for Indo-European languages to have dependentmarking rather

than head marking. Subject–verb agreement is illustrated by the Latin sen-

tence (11): the form lavat agrees in number andpersonwith the subjectmanus.

Head marking in Indo-European languages can also be illustrated by

quantiWers that denote a quantity higher than 1. These quantiWers require

the plural form of the head noun if it is countable, as in English two books.

This is a case of government in which the number of nouns (normally a case

of inherent inXection) plays a role in contextual inXection. In the non-

Indo-European language Hungarian, on the other hand, the noun in such

phrases is forbidden to have a plural marking, and the whole noun phrase

behaves as singular with respect to verb agreement (Corbett 2000: 211):

(21) Ket leny besze

two girl.sg chat.sg

‘‘two girls are chatting’’

In several Slavonic languages, numerals require speciWcmorphologicalmark-

ings on their nouns. For instance, theRussian numerals for 2, 3, and 4 require

the nouns with which they combine to appear in the genitive singular form.

In some languages the verb is not only marked for the properties of its

subject, but also for its dependent NPs such as the accusative or dative

object, as illustrated here for the Austronesian language Kambera spoken

on the island of Sumba, Indonesia (Klamer 1998: 63):

(22) a. (Na tau wutu) na-palu-ka (nyungga)

the person be.fat 3sg.nom-hit-1sg.acc I

‘‘the big man hit me’’

b. (I ama) na-kei-nya ri

art father 3sg.nom-buy-3pl.dat vegetable

‘‘Father buys them vegetables’’

The parts of these sentences that are between parentheses can be omitted.

Therefore, the aYxes on the verb are best interpreted as pronominal aYxes

110 inflection

that indicate the subject and the object of the verb. These aYxal pronouns

can then be said to have a relation of co-reference with the NPs in the

sentence. In other words, you might paraphrase a sentence such as (22a)

as ‘As to the big man and me, he hit me’. Such sentences are therefore

comparable to an English sentence such as John, he hit it, the ball.

In a number of languages with subject–verb agreement, the subject with

which the verb is supposed to agree in person and number can be absent. For

instance, the Latin verb form laudat in (2) can function as a sentence by itself,

with the meaning ‘‘he praises’’. Thus, the verb form suYces to identify the

number and person of the subject of a clause. That is, the choice of one and

the same inXectional form is sometimes determined by syntactic context, as in

Carolus laudat ‘‘Charles praises’’ (where the verbal form has to agree in

person and number with the subject Carolus), whereas in other cases that

choice is not required by another element in the syntactic context, as in the

sentenceLaudat ‘‘He praises’’. The same observation can be made for gender

agreement in French. In this language, predicatively used adjectives must

agree in gender with the subject, as illustrated in the following sentences:

(23) a. Mon mari est heureux

My.masc.sg husband.masc.sg is happy.masc.sg

b. Ma tante est heureus-e

My.fem.sg aunt.fem.sg is happy-fem.sg

c. Je suis heureux

I am happy.sg.masc

d. Je suis heureus-e

I am happy-sg.fem

In (23a, b) the features of the subject are imposed on the adjective. In (23c, d)

on the other hand, the pronoun je ‘‘I’’ has no inherent features for gender,

and the gender expressed on the adjective reXects the gender pragmatically

assigned to the subject je. These facts suggest that agreement should not be

interpreted as the transfer of certain features from one word or constituent

to another, but as a checking device that checks if the features of the

constituents in the relevant syntactic conWguration do not contradict each

other. In (23a, b) the features are identical, and hence non-contradictory.

In (23c, d) the features of je are non-contradictory (though not identical) to

those of the predicate: je has no speciWcation for gender, whereas the

predicative adjectives do have such a speciWcation.

An eVect of agreement is to indicate which words belong together in a

phrase. This may have the eVect that constituent words of a phrase with

inflection 111

agreement can be separated by other words, as is frequently the case in

Latin texts. In the following example, the complement mundi separates the

adjective and the head noun with which it agrees in gender, number and

case (Virgil, Georgics 1.5):

(24) O clar-issim-a mund-i lumin-a

O clear-superl-neut.pl.voc world-gen.sg light-neut.pl.voc

‘‘O, clearest lights of the earth’’

Australian languages such as Warlpiri are also well-known for this kind of

freedom in word order, made possible by the case-marking system. Both case

marking and agreement patternsmake it possible for a language to havemore

or less free word order because theymay serve to Wnd out whichwords belong

together, and which function they have. However, this is not an automatic

consequence: the presence of rich inXection does not necessarily imply that a

language has free word order. German, for instance, has pretty rigid word

order principles notwithstanding its rich case and agreement system.

Contextual inXection introduces a large degree of redundancy, and it may

therefore come as no surprise that contextual inXection erodes much more

frequently in the course of history than inherent inXection. The Romance

languages French, Italian, and Spanish, descendants of Latin, all kept the

number distinction for nouns, whereas these languages no longer have case

markings on nouns. Afrikaans has lost its verbal endings for number and

person(contextual inXection),whereas ithaskeptthetensedistinctionbetween

present and past. This shows that the distinction between inherent and con-

textual inXection is important for understanding patterns of language change.

5.3 Inflection and derivation

As we saw in Chapter 1, the primary distinction between inXection and

derivation is a functional one: derivation creates new lexemes, and inXec-

tion serves to create diVerent forms of the same lexeme. Yet, this is not

always suYcient to determine in concrete cases of morphology to which

domain a particular morphological form belongs. Consider English com-

paratives. How do we know if bigger is a diVerent lexeme than big, or

another form of the lexeme big?

We might deWne inXection as ‘the kind of morphology that is relevant

to syntax’. According to that demarcation criterion, the morphological

112 inflection

properties that play a role in agreement and government are clear cases of

inXection. This comprises all contextual inXection, but also those morpho-

logical properties of words that function as controllers for this kind

of inXection. The marking of number on nouns is often not an instance of

contextual inXection itself, but it may play a role in determining the shape of

adjectives and determiners with which it combines. Note, however, that we

cannot say that derivation is completely irrelevant to syntax. For example,

when we create causative verbs by means of derivation, we create transitive

verbs, and transitivity is certainly a property that is relevant to syntax.

A second possible criterion is that inXection is obligatory, whereas deriv-

ation is optional. This criterion does apply to contextual inXection, but at Wrst

sight not always to inherent inXection. In the case of verbal conjugation,

inXection is always obligatory: you have to choose a speciWc form of a verb

in a clause. This seems not to apply to nouns: a noun can be used without any

morphological marking for number. In fact, for many nouns the need for a

plural form will (almost) never arise, as is the case for the English nouns

attention, accordance, and adolescence. However, onemay claim that English

words are always inXected for the relevant categories because anEnglishnoun

is always either singular or plural. After all, these latter three nouns behave as

singular nouns in subject–verb agreement. So these nouns are singular ‘‘by

default’’. In this sense, inXection for number is indeed obligatory in English.

An important criterion that might distinguish inXection from derivation

is the essential role of the paradigm in inXection. The cells of the paradigm

are deWned by the inXectional categories of a particular word class. In

Chapter 6 the role of paradigms in making morphological generalizations

is discussed. A clear example of the role of paradigms can be found

in periphrasis. We have to do with periphrasis if for certain cells of the

paradigm no syntheticmorphological form is available. Instead, a combin-

ation of words, an analytic or periphrastic form, has to be used. Latin has no

synthetic forms for the perfective passive of verbs, as illustrated in Table 5.2

for the 3sg forms of laudare ‘‘to praise’’. The cells for the perfective passive

are a combination of the passive participle (that, like adjectives, agrees with

the subject of the clause with respect to case, gender, and number) and

a form of the verb esse ‘‘to be’’. If these word combinations were not

considered part of the verbal paradigm, Latin verbs would have a paradigm

with a gap for the perfective passive forms. These periphrastic forms have a

perfective interpretation, although the forms of the verb esse ‘‘to be’’ are

that of the imperfect tense.

inflection 113

An additional argument for considering these word combinations as

Wlling paradigm cells is the following. Latin has a number of so-called

deponent verbs, verbs with a passive form but an active meaning. For

instance, the verb loquor ‘‘to speak’’ is such a deponent verb. The crucial

observation is that a word-sequence such as locutus est receives an

active interpretation as well, and means ‘‘he has spoken’’. This parallelism

in interpretation as active forms is to be expected if these analytic forms

belong to the inXectional paradigm of verbs.

The notion ‘suppletion’ also presupposes the idea of a paradigm. We

speak about the grammatical words am, are, is, was, and were as forms of

the English lexeme be although they are quite diVerent in phonological

shape, and show (almost) no phonological relatedness. These words Wll

speciWc cells in the paradigm of to be. The same applies to worse, the

suppletive comparative form of bad.

A fourth criterion for distinguishing inXection and derivation is that

derivation may feed inXection, but not vice versa. Derivation applies to

the stem-forms of words, without their inXectional endings, and creates

new, more complex stems to which inXectional rules can be applied. This is

the main reason for keeping the two kinds of morphology distinct. It is a

cross-linguistic generalization that inXection is peripheral with respect to

derivation, formulated by Greenberg as follows:

(25) ‘Universal 28. If both the derivation and inXection follow the root, or they

both precede the root, the derivation is always between the root and the

inXection’ (Greenberg 1963:93).

Greenberg’s universal excludes the morpheme order patterns *Derivation–

InXection–Root and *Root–InXection–Derivation. It might also be

Table 5.2. Imperfective and perfective 3 sg forms of laudare

imperfective active passive

present laudat laudatur

past laudabat laudabatur

future laudabit laudabitur

perfective active passive

present laudavit laudatus/a/um est

past laudaverat laudatus/a/um erat

future laudaverit laudatus/a/um erit

114 inflection

interpreted as saying that derivation cannot apply to inXected forms. Yet,

there are exceptions to this universal tendency, cases in which inXectional

forms appear to feed derivation. For instance, the comparative form of

some Dutch adjectives has functioned as the base for derivation with the

preWx ver-, as in

(26) erg-er ‘‘worse’’ [ver-[erg-er]A]V ‘‘to worsen’’

oud-er ‘‘older’’ [ver-[oud-er]A]V ‘‘to get older’’

Strictly speaking, this morphological pattern is not excluded by Green-

berg’s Universal 28 because the derivational morpheme is a preWx, and the

inXectional morpheme is a suYx. A real exception to this universal is the use

of verbal participles (which may function as adjectives) for de-adjectival

word-formation, which is quite common in Indo-European languages. For

example, English past participles show up in de-adjectival word-formation,

as in aVect-ed-ness and relat-ed-ness. Here we Wnd the ‘‘wrong’’ order Root–

InXection–Derivation, since -ed is an inXectional suYx, and -ness a deriv-

ational one. However, this kind of de-inXectional word-formation is only

found with instances of inherent inXection such as comparatives and

participles as bases.

Another demarcation criterion that might be invoked is that derivation is

potentially category-changing, unlike inXection. Although it is true that

most cases of inXection do not change syntactic category, there is a change

of category involved for inWnitives, gerunds, participles, and converbs,

which keep their verbal potential, but also have properties of other syntac-

tic categories (section 5.1).

In sum, the best criteria for distinguishing inXection from derivation are

the obligatoriness of inXection, the fact that it is organized by means of

paradigms, and that it is normally a word without its inXectional endings

(¼ the stem) that forms the basis for word-formation. It will be clear that

the boundary between the two is not extremely sharp, and that there are

similarities between inherent inXection and derivation.

5.4 Theoretical models

The inXectional phenomena discussed above pose two speciWc questions

for the theory of grammar: (i) what is the best formal representation of

inXectional processes, and (ii) where in the grammar should inXectional

rules apply?

inflection 115

Let us Wrst focus on the formal nature of inXectional rules. For simple

cases, one might think of inXection as the attachment of inXectional mor-

phemes to the stem-forms of lexemes. For instance, in English we create

plural forms of nouns by suYxing the stem with the suYx -s. Similarly, past

tense forms of verbs are made by suYxation of the verbal stem with -ed.

Such cases of agglutinative morphology can therefore be dealt with in a

model in which morphology is seen as the concatenation of morphemes.

This model is called Item-and-Arrangement Morphology.

There are two basic problems for this model of inXection. The Wrst is that

in many languages there is no one-to-one relation between inXectional

properties and their expression by morphemes. Consider once more the

paradigm of the Polish noun kot in Chapter 2, repeated as (27) here for


nominative kot kot-y

genitive kot-a kot-ow

dative kot-u kot-om

accusative kot-a kot-y

instrumental kot-em kot-ami

locative koci-e kot-ach

vocative koci-e kot-y

convenience. Each inXectional suYx in this paradigm expresses features (is

an exponent) for two categories, number and case. There are no distinct

morphs for these two categories, and the inXectional endings are portman-

teau morphs. This is a case of cumulative exponence: each ending in the

paradigm of kot is the expression of more than one inXectional category

(the formal correlate of a morphological category is called its exponence).

There is also the opposite phenomenon that one inXectional category may

receive more than one morphological expression. This is illustrated by the

Latin word form laudavisti in Table 5.1: the inXectional category perfect is

expressed by both the morpheme -vi- after the stem lauda and the selection

of a 2sg ending -isti that is unique to the perfect, and hence also expresses

this category. This is called extended exponence.

The second problem for an Item-and-Arrangement model of inXection

is that inXectional categories may be expressed by other means than mor-

pheme concatenation. In Germanic languages, for instance, the past tense

116 inflection

forms of so-called strong verbs are formed by changing the vowel of the

verbal stem (ablaut), not by suYxation.

These problems for the Item-and-Arrangement model have led to a diVer-

ent view of inXection in which inXectional rules are operations of various

sorts: aYxation, vowel change, reduplication, etc. This process-variant

of morphology is called Item-and-Process Morphology. Thus, the rule for

computing the pl.instr. form of Polish nouns like kot might be formulated

as follows:

(28) [x]N ! [x-ami]N[þpl:,instr]

This rule states that if we add the ending -ami to the stem form of a noun,

we thus create a noun with the features plural and instrumental.

A third model for inXection, which shares the processual view of morph-

ology with the model of Item-and-Process morphology, is the Word-and-

Paradigm model. This model takes the lexeme and its paradigm of cells as

its starting point. The diVerent forms of the paradigm of a lexeme are

computed by a set of realization rules. The realization rule for the instr.pl.

form of nouns like kot will have the following form:

(29) [x]N [þpl, instr] ! [x-ami]N

The diVerence between rule (28) and rule (29) is the following. Rule (28)

introduces a morphosyntactic property hand-in-hand with its exponent,

while rule (29) treats the property set as a precondition for the introduction

of its exponent. Rules such as (29) are therefore called realization rules or

spell out rules, and this kind of morphological analysis is called realizational

morphology.

One advantage of using the format of spell out rules for inXection is that

it can easily account for cases in which there is no overt phonological

expression for certain morphosyntactic features. In English and Dutch,

there is no overt ending for singular nouns. Yet, nouns must be speciWed for

number in order for agreement rules to apply properly. One might there-

fore assume a zero suYx or preWx for singular nouns, but this is an ad hoc

assumption. It is even impossible to determine if the zero is a preWx or a

suYx, since you cannot hear it. In a realization approach one can account

for this straightforwardly. Since the cell with the feature [� plural] of a

Dutch or English noun will not trigger a realization rule, the singular form

of the noun will be identical to the stem-form. This makes the introduction

of a zero aYx superXuous.

inflection 117

The basic analytical problem posed by inXectional systems discussed

above is that there is no neat mapping of form and (morphosyntactic)

content in a one-to-one-fashion. A related problem is that there is quite

often competition between diVerent realization rules for the same morpho-

syntactic content. In Germanic languages, for instance, the past tense form

of verbs is created either by vowel change, or by suYxation. In the case of

sing, with the past tense form sang, we have to take care that the grammar

will not specify its past tense form as singed. This will be achieved as

follows. The rule that applies to ablauting verbs such as sing will not only

be conditioned by the presence of the feature past tense, but also by the

diacritic feature [ablaut] that is assigned to these ablauting verbs in the

lexicon. A diacritic feature refers to an arbitrary class of lexical items. From

the synchronic point of view it is an arbitrary property of a verb that it

belongs to the class of verbs that makes use of vowel alternations to express

past tense. The additional presence of the feature [ablaut] in the condition

for vowel change makes this rule more speciWc than the rule for suYxation

with -ed, which only requires its inputs to be verbs. It is a generally accepted

idea in linguistics that, in the case of competition between two or more

rules, the more speciWc rule will be applied Wrst, and then pre-empts the

application of the more general rule. This is called Panini’s Principle, after

the Sanskrit grammarian Panini who introduced this idea.

Another theoretical model that should be mentioned here is that of

Distributed Morphology. This model is not process-based, but morpheme-

based: morphemes are the atoms of morphosyntactic representations, and

to that extent it is a (sophisticated) variant of Item-and-Arrangement

Morphology. At the syntactic level, the structure of a sentence is repre-

sented as a syntactic tree that includes abstract morphemes such as

[þplural], [þpast], etc. These abstract morphemes form terminal nodes in

the syntactic tree, just like lexical morphemes. The phonological content of

these morphemes is then spelt out through the insertion of vocabulary

items that specify the phonological correlate of an abstract morpheme.

An example of a vocabulary item is the English plural suYx /z/ for nouns

(as in dog–dogs):

(30) =z= $ [ _ ,þplural]

This vocabulary item states that the phonological piece /z/ can be inserted

in the context of the feature [þplural]. In this approach, there is no lexicon

that provides full words, to be used in syntactic structure. The task of

118 inflection

providing fully inXected words is distributed over other components of the

grammar, hence the name Distributed Morphology.

How does the model of Distributed Morphology deal with cumulative

exponence, the phenomenon that poses a problem for Item-and-Arrange-

ment Morphology? In such situations, in which one phonological piece

expresses a set of morphosyntactic features, the morphosyntactic tree will

be restructuredbefore lexical insertion takes place. For instance, the abstract

morphemes for number and case in Latin and Polish may be fused into one

terminal node through a fusion rule. Subsequently, the vocabulary item that

matches with that feature combination is inserted, for instance the Latin

suYx -i for the nom. pl form of second declension nouns:

(31) =i=$ [ _ ,þplural,þnom]N, class II

In English plural nouns such as geese and men, the plurality is expressed by

vowel change, not by the addition of a morpheme. In Distributed Morph-

ology, this is taken care of by assuming a zero suYx for these plural nouns,

followed by the application of readjustment rules that perform phono-

logical operations triggered by the presence of the feature [þplural] for a

speciWed number of lexical items.

5.5 Morpheme order

The issue of morpheme order deserves special attention in the domain of

inXection. In the case of derivation the basic idea is that each morpho-

logical operation adds a new layer of structure, and a corresponding

semantic interpretation, as was illustrated in section 1.2 for the word tran-

quilizer, the morphological structure of which could be represented by

means of a hierarchical structure. In the domain of inXection, however, a

Xat structure might be more appropriate. When there are diVerent endings

for tense and for person, with the second ending being more peripheral,

there may be no particular reason for assuming a hierarchy in which

person is higher than tense. This is why some linguists assume that with

respect to inXection we have to do with a set of unordered features, and

why the expression ‘‘set of morphosyntactic features’’ is used. The order in

which inXectional elements appear is sometimes expressed by means

of templates. Well-known examples of languages with such templatic

morphology (also called position class morphology) are the Bantu languages.

inflection 119

For instance, the following (partial) template can be assumed for Bemba, a

Bantu language of Zambia:

(32) Negation—Subject marker—Tense—Aspect—Object marker—Stem—Final

vowel

This template is exempliWed by the following sentence (Kula 2002: 33):

(33) ta- tu- aku- laa- ba- bombel -a

neg subj.1pl fut progr obj work final vowel

‘‘We will not be working for them’’

The assumption of templates does not mean, however, that morpheme

order is completely arbitrary, and that there are no tendencies in the

ordering of inXectional morphemes. One clear generalization is that con-

textual inXection is peripheral to inherent inXection. Thus, in the case of

nouns, aYxes for case are peripheral to aYxes for number (Greenberg’s

Universal 39, Greenberg 1963: 95), and in the case of verbs person endings

tend to be peripheral to tense endings. A second generalization concerns

the order of morphemes of inherent inXection such as those for tense and

aspect. Generally, aspect morphemes appear closer to the root of a word

than tense morphemes (Bybee 1985).

If word-formation and inXection are indeed diVerent subsystems of

morphology, does this imply that they are located in diVerent parts of the

grammar? This is the second main issue for theoretical models of inXection:

where in the grammar is inXection?

One possible view is that word-formation is pre-syntactic, and inXection

is post-syntactic. Word-formation (compounding, derivation, etc.) serves

to enlarge the set of lexical items that can be inserted into syntactic

structure. InXection, on the other hand, may be claimed to be post-syntactic

because the speciWc form of a lexeme may depend on its syntactic context

(contextual inXection). This is the split morphology model:

(34) word-formation ! syntax ! inXection

However, the dependency of inXection on syntax does not necessarily

imply that inXection is post-syntactic. The alternative view is that the

morphological component of the grammar computes new lexemes and

the diVerent inXectional forms of lexemes, the theory of strong lexicalism:

(35) morphology ! syntax

120 inflection

Rules of contextual inXection like those of agreement will then have the

function of checking mechanisms: they check if the morphosyntactic fea-

tures of the words in a particular syntactic conWguration can go together.

If not, the sentence is qualiWed as ungrammatical. For instance, if we insert

the word form books with the feature [þ plural] in the context—is interest-

ing, we get the ungrammatical sentence *Books is interesting. The English

rule of subject–verb agreement will Wnd a clash between the number

speciWcations of books and is, and hence, the grammar will qualify it as

an ungrammatical sentence. Moreover, as we saw above, the choice of a

particular inXected form like the French fem.sg adjective heureuse does not

always depend on the presence of a syntactic source for the feature [femi-

nine]. In Je suis heureuse ‘‘I am happy’’, this form is selected because the

speaker is female. Hence, it cannot be created post-syntactically through

agreement with the gender features of the subject, because the subject je is

unspeciWed for gender.

Another advantage of the strong lexicalist position is that we can

account for the fact that certain kinds of (inherent) inXection feed word-

formation. For instance, in many languages participles may function as

adjectives, and feed de-adjectival word-formation (as in English aVected-

ness). In the split morphology model it is impossible for inXectional forms

to be used as bases for word-formation since inXection is post-syntactic,

whereas word-formation comes before syntax.

An additional consideration for dealing with word-formation and inXec-

tion in the same component of the grammar is that derived words and

inXected forms of words may be subject to the same phonological rules or

constraints. These phonological regularities within words are referred to as

the lexical phonology of a language. For instance, all Dutch words are

subject to the rule that the vowel schwa [

e

] disappears before an adjacent

vowel. In the case of vowel-initial suYxes that trigger this schwa-deletion, it

does not matter if the suYx is inXectional or derivational in nature:

(36) derivation zijde-ig ‘‘silky’’ /zEØd

e

-

eg/ [zEid e

x]

inXection bode-en ‘‘messengers’’ /bo:d

e

-

e

n/ [bo:d

e

n]

The rule of schwa-deletion should not apply twice, both pre-syntactically

and post-syntactically, because that would complicate the grammar of

Dutch beyond necessity. This duplication would be necessary if we followed

the theory of split morphology, unless all phonology were post-syntactic.

In Chapter 7 it is shown that this latter position cannot be correct.

inflection 121

These considerations lead to the conclusion that morphology should not

be split into a pre-syntactic and a post-syntactic component, and that it is

only the strong lexicalist position that can do justice to the facts discussed

here.

Summary

InXection is the expression of morphosyntactic properties of lexemes. These

properties either serve to express a particular meaning (inherent inXection),

or are required in speciWc syntactic contexts (contextual inXection). In

contextual inXection the relation between two elements in a syntactic con-

Wguration is marked either on the head or on the dependent. Contextual

inXection indicates syntactic relationships between words. Word order also

has the function to indicate these relationships. Hence, a language with a

rich system of contextual inXection may have relatively free word order.

The main criterion for distinguishing inXection from derivation is that

only inXection is obligatory: each word must be speciWed for the relevant

inXectional properties of its word class. Each lexeme thus has a paradigm of

inXectional forms. It is normally the word without its inXectional endings

that forms the starting point for word-formation. However, certain types

of inherent inXection may feed derivation.

The complications in the relation between phonological form and mor-

phosyntactic properties have given rise to a number of theoretical models

that are alternatives to the Item-and-Arrangement model that analyses

(inXectional) morphology as morpheme concatenation. These alternative

models are processual models of inXection.

Questions

1. Give an explanation for the fact that the German phrase Guten Morgen‘‘good morning’’ has the accusative form although there is no verb thatassigns accusative case to this phrase.

2. What are the rules for choosing between a synthetic and a periphrasticexpression of the comparative of the English adjective?

3. In English, the infinitival form of the verb is identical to that of the stem.Is it therefore necessary to assume a zero ending for this form?

122 inflection

4. In Dutch we find compound pairs such as the following: school-gemeenschap‘‘school community’’ versus schol-en-gemeenschap ‘‘schools community,comprehensive school’’, and stads-raad ‘‘city council’’ versus sted-en-raad‘‘cities’ council’’ (-en is a plural morpheme). Which conclusion can you drawfrom the existence of such compounds about the relation between inflectionand word-formation?

5. Which kinds of inherent and contextual inflection can be observed in thefollowing Finnish sentence?

Anno-i-n vet-ta kahde-lle koira-llegive-IMPF-1 SG water-PART two-ALL dog-ALL‘‘I gave water to two dogs’’

6. Melcuk (2000: 515) gives the following examples of Russian male–femalenoun pairs that he qualifies as suppletion: byk ‘‘bull’’ –korova ‘‘cow’’, petuch‘‘rooster’’–kurica ‘‘hen’’. Do you think it makes sense to use the notion‘‘suppletion’’ in the domain of word-formation?

7. Identify instances of cumulative and extended exponence in the followingforms of the Hebridian Gaelic word clach ‘‘stone’’ (Coates 2000: 623):

SG PL

NOM clach [khlax] clachan [klaxan]GEN cloiche [khloce] clachan [klaxan]DAT cloich [khloc] clachan [klaxan]

8. Consider the following sentences from British English and American Englishrespectively:

The committee are investigating the caseThe committee is investigating the case

What conclusion can you draw for the difference in specification of committeebetween the lexicons of British and of American English?

9. French has two genders, masculine and feminine. The following Frenchnoun phrases illustrate gender agreement between the head noun and thefollowing attributive adjective:

un pere excellenta.MASC.SG father.MASC.SG excellent.MASC.SG

‘‘an excellent father’’une mere excellent-ea.FEM.SG mother.FEM.SG excellent-FEM.SG

‘‘an excellent mother’’une mere et une sœur excellent-e-sa.FEM.SG mother.FEM.SG and a.FEM.SG sister.FEM.SG excellent-FEM-PL‘‘an excellent mother and sister’’un pere et une mere excellent-sa father.MASC.SG and a mother.FEM.SG excellent-MASC.PL

‘‘an excellent father and mother’’

inflection 123

Formulate the rule of ‘gender resolution’ that is needed to compute thegender feature of the attributive adjective when the head of the phrase isformed by two conjoined nouns.

10. Dutch has fifteen nouns with a plural suffix -eren instead of the regular -en,for example kind-kinderen ‘‘child-children’’. The regular suffix for mono-syllabic nouns is -en, as in boek-boeken ‘‘book(s)’’. Which principle can beinvoked to block the formation of *kinden?

Further reading

The distinction between inherent and contextual inflection has been made inAnderson (1985) as the distinction between inherent and relational inflection;the importance of this distinction for morphological theory is discussed in Booij(1994, 1996a). Panini’s principle is discussed in Pinker (1999).The Word-and-Paradigm model has been defended in a classic article by

Robins (1959), and in Matthews (1972). A survey and classification of inflec-tional theories is given in Stump (2001: ch 1). Realization rules for inflection aredefended in Anderson (1992) and Stump (2001), which are detailed studieswithin the Word-and-Paradigm approach. Borjars et al. (1997) and Sadler andSpencer (2001) deal with the theoretical status of periphrasis.Key publications on the model of Distributed Morphology are Halle (1992)

and Halle and Marantz (1993). A survey of the model is presented in Harleyand Noyer (2003).The split morphology model is defended in Anderson (1992). Objections to

this model are raised in Booij (1994, 1996a).Templatic morphology is discussed in Inkelas (1993), Rice (2000), and

Hyman (2003).

124 inflection

6Inflectional systems

6.1 Nominal systems: gender, number, and case

The previous chapter has introduced you to the roles of inXection in con-

structing sentences, and its position in the grammar. In this chapter, we will

zoom in on the details of inXection. The formal regularities involved are

sometimes quite complicated. Moreover, the relationship between inXec-

tional forms and their semantic interpretation is not that straight-forward.

Thus, inXectional phenomena give a perfect illustration of how complex the

relation between form and meaning in natural languages can be.

Let us Wrst have a more detailed look at how inXection manifests itself in

the forms of nouns. The best known cases of noun inXection are the

marking on nouns of number and case properties. In an agglutinative

language such as Finnish, number and case are marked by means of

separate suYxes. In most languages, case markings are external to number

markings, in line with the generalization that contextual inXection tends to

be peripheral to inherent inXection. In Finnish, nouns have either singular

or plural number. As in many other languages, there is no overt marking

for the singular. The plural suYx is -t in the nominative form and -i

elsewhere. The latter suYx is realized as the glide [ j ] when it appears

between two vowels. A subset of plural case forms of the Finnish noun

kissa ‘‘cat’’ is given in (1) (Sulkala and Karjalainen 1992: 267):

6.1 Nominal systems:

gender, number, and case 125

6.2 Categories of verbal

inXection 133

6.3 Autonomous morphology 141

Summary 146

Questions 146

Further reading 149

(1) nom.pl kissa-t

gen.pl kisso-j-en

part.pl kisso-j-a

abl.pl kisso-i-lta

This example illustrates that both stems and aYxes in inXectional

forms may exhibit phonological variation of various kinds, a topic that is

discussed in Chapter 7.

The number system of English nouns has two characteristic features:

number marking is obligatory, and the only choice that we have is that

between singular and plural. For many languages, expression of number on

nouns is not obligatory. This is the case for Austronesian languages such as

Malay. Such languages have a general form for nouns which is unspeciWed

as to number, and stands outside the number system.Other languagesmight

express number obligatorily, but not on nouns. In Maori, for instance, it

is expressed on possessive markers and determiners only. Languages may

have more distinctions than just singular versus plural. In fact, cross-

linguistically, number marking is not restricted to the distinction between

singular and plural. Some languages also distinguish a dual form, referring

to precisely two entities, a trial form (three entities), or a paucal form

(referring to a few entities). There is a cross-linguistic hierarchy involved

here (Corbett 2000: 38):

(2) Number hierarchy: singular > plural > dual > trial

This hierarchy indicates that, if a language has a trial, it also has a dual and

if it has a dual, it also has a plural form, etc. The dual and trial may be

diVerent from the plural in not being obligatory. That is, in some cases, the

plural can be used for denoting a set of two and three entities respectively,

in addition to larger numbers of entities.

Bayso, a Cushitic language spoken in southern Ethiopia, exhibits optional

number marking: a noun can be usedwithout any overt marking for number,

and then receives a general interpretation, as shown in the Wrst example in (3).

This language features singular, plural, andpaucalnumber (Corbett 2000:11):

(3) luban foofe

lion.general watched.1sg

‘‘I watched lion’’ (one or more than one)

luban-titi foofe

lion-sg watched.1sg

‘‘I watched a lion’’

126 inflection

luban-jaa foofe

lion-paucal watched.1sg

‘‘I watched a few lions’’

luban-jool foofe

lion-pl watched.1sg

‘‘I watched (a lot of) lions’’

When a language has a number distinction for nouns, this is not necessarily

expressed on all nouns. In this respect, number marking is certainly not the

inXectional category par excellence, since we consider obligatoriness as

a relevant criterion for the demarcation of inXection and derivation

(section 5.3).

A second typological dimension of number marking is that languages

may diVer as to which nouns are marked for number. A language may

mark number on animate nouns only, or only on nouns that denote human

beings. The cross-linguistic variation involved obeys theAnimacyHierarchy

(Corbett 2000: 56):

(4) speaker > addressee > 3pers > kin > human > animate > inanimate

Each language can be characterized by a particular cut-oV point on this

hierarchy for number marking. For instance, it may be the case that in a

certain language only personal pronouns (all three persons) and nouns

denoting kinship relations are marked for number. This hierarchy can be

used to formulate certain constraints on the cross-linguistic variation found.

For instance, if the singular–plural distinction is made for animate nouns, it

will also be made for kinship nouns (but not necessarily vice versa).

In English, there is no overt morphological marking for singular, only

for plural. In other languages both singular and plural may be expressed

phonologically. The inverse situation, no overt marking for plurals, and

overt marking for singular forms is also found.

Number marking on nouns, although primarily a case of inherent inXec-

tion, may also be required by a speciWc syntactic context, in particular in

combination with number names. In English we must use the plural form of

nouns after number names higher than 1. In Finnish and Hungarian, on the

other hand, the singular form of a noun is required after such number

names.

Through agreement, number can also be marked on words of other word

classes such as adjectives, determiners, and verbs, as discussed in section

5.2. This kind of number marking is to be distinguished from inherent

inflectional systems 127

marking of number on verbs used to express the plurality of the event

denoted by the verbs, or the number of participants involved. The follow-

ing examples are from Ngiti (Kutsch Lojenga 1994: 285, transcription

simpliWed):

(5) ma m-i indri nadha

1sg sc-aux goat pull.nom1

‘‘I am pulling one goat, or a group of goats simultaneously’’

ma m-i indri nudha

1sg sc-aux goat pull.pl.nom1

‘‘I am pulling several goats one by one’’

(The plural verbal form here diVers from the unmarked form by having a

diVerent vowel.) This plural verbal form denotes that the action takes place

more than one time.

In languages with diVerent inXectional classes such as Latin (see Chapter 5,

example 9), number and case receive diVerent morphological markings

dependent on the inXectional class to which a noun belongs. This is also

illustrated by the inXectional paradigms of some Icelandic nouns of class

I (the default class) in Table 6.1. In these paradigms, three classiWcations of

nouns are involved in computing the form for each number–case combin-

ation: the distinction between weak and strong nouns, gender, and arbitrary

class markers. The words in Table 6.1 are all class I nouns, but there are three

Table 6.1. InXection of Icelandic class I nouns

masculine feminine neuter

strong weak strong weak strong weak

‘‘horse’’ ‘‘time’’ ‘‘needle’’ ‘‘tongue’’ ‘‘table’’ ‘‘eye’’

singular

nom hest-ur tım-i nal tung-a borð aug-a

acc hest tım-a nal tung-u borð aug-a

dat hest-i tım-a nal tung-u borð-i aug-a

gen hest-s tım-a nal-ar tung-u borð-s aug-a

plural

nom hest-ar tım-ar nal-ar tung-ur borð aug-u

acc hest-a tım-a nal-ar tung-ur borð aug-u

dat hest-um tım-um nal-um tung-um borð-um aug-um

gen hest-a tım-a nal-a tung-na borð-a aug-na

Source: Thrainsson 1994: 153.

128 inflection

other classes (for non-neuter nouns only) and some irregular nouns. The

phonological form of a noun may help to determine the (sub)class of a

noun. For instance, if the nom.sg form of a masculine noun ends in -i (as

in timi ‘‘time’’), it always follows the weak inXection pattern.

A closer look at the paradigms in Table 6.1 will reveal that the number of

diVerences between the paradigms of Icelandic class I nouns is restricted.

For instance, the dative plural marker is always -um, and the genitive

plural always ends in -a (with an additional n in tung-na and aug-na).

Another generalization is that all feminine and neuter nouns have the

same form for the nominative and the accusative, in the plural. That is,

the accusative forms can be computed by referring to the nominative

forms. To give another example of such paradigm regularities, in Russian

the accusative form is identical to the genitive form for most animate

nouns, and for inanimates, the accusative is like the nominative (Corbett

and Fraser 1993: 130). Such generalizations that refer to diVerent cells of a

paradigm are called rules of referral since one form can be computed by

referring to another form of the same paradigm. It is important to note that

rules of referral do not refer to word forms, but to diVerent cells in

the paradigm of a word. For instance, the following regularity holds

for the masculine nouns in Table 6.1: if the nominative plural form ends

in -ar, the accusative plural and the genitive plural end in -a. Rules of

referral underscore the important role of the paradigm as a pattern of cells,

each with it is own set of morpho-syntactic features.

When diVerent slots of a paradigm are Wlled by the same phonological

form, this is referred to as syncretism or inXectional homonymy. In Table 6.1

we see that the nominative and accusative plural forms of feminine and

neuter nouns of class I are always identical. This is a case of systematic

syncretism that may be expressed by the following rule of referral:

(6) [Noun, Class 1, Non-masc] ! [Nom ¼ Acc in Plural]

Rules of referral can thus be used to express systematic patterns of sync-

retism.

The classiWcation of nouns into diVerent genders is quite an intriguing

phenomenon because of its strong arbitrariness. Some Indo-European

languages have three genders: masculine, feminine, and neuter. Romance

languages such as Italian and French have lost the neuter gender, whereas

in the Germanic language Dutch there is only a distinction between neuter

and non-neuter gender (also called common gender). There is a relation


between gender and biological sex, to be sure. If a language distin-

guishes between masculine and feminine gender, morphologically simplex

nouns denoting male and female beings tend to be masculine and feminine

respectively. But why are some nouns for body parts in Dutch neuter

(for instance, oog ‘‘eye’’), whereas others are non-neuter (for instance

neus ‘‘nose’’)? Linguists assume that originally there must have been some

semantic motivation behind the diVerent classes, a motivation that became

opaque in the course of history.

Morphological structure may also be involved in computing the gender

of a noun. The heads of nominal compounds, and most suYxes determine

the gender of the words they create. As we observed earlier, Dutch and

German diminutive nouns, for instance, are always neuter, even if they

denote male or female beings:

(7) Dutch het jonge-tje ‘‘the boy’’; German das Mutter-chen ‘‘the little mother’’

Another system of noun classes involved in the morphological expres-

sion of number is found in languages of Africa, in particular (but not

exclusively) Bantu languages. The examples in Table 6.2 illustrate the noun-

class system of Swahili. These data suggest that Swahili has six classes

(genders), each class being a particular pair of singular and plural forms.

The Wrst two singular (sub)classes look as if they are one class because

they have the same preWx m-. However, they correlate with two diVerent

plural preWxes. Moreover, they also behave diVerently in agreement.

This is illustrated by the pattern of subject agreement preWxes in the verbal

forms in (8), in which the nouns listed in Table 6.2 function as subjects,

and are followed by the verbal form mepotea ‘‘is/are lost’’ (Welmers

1973: 162):

Table 6.2. Noun classes in Swahili

singular plural

I m-tu wa-tu ‘‘person’’

II m-zigo mi-zigo ‘‘load’’

III Ø-tofali ma-tofali ‘‘brick’’

IV ki-tasa vi-tasa ‘‘lock’’

V n-dizi n-dizi ‘‘banana’’

VI u-bao m-bao ‘‘plank’’

Source: Welmers 1973: 161.

130 inflection

(8) singular plural

m-tu a-mepotea wa-tu wa-mepotea

m-zigo u-mepotea mi-zigo i-mepotea

Ø-tofali li-mepotea ma-tofali ya-mepotea

ki-tasa ki-mepotea vi-tasa vi-mepotea

n-dizi i-mepotea n-dizi zi-mepotea

u-bao u-mepotea m-bao zi-mepotea

The singular nouns m-tu and m-zigo select diVerent subject agreement

preWxes. Therefore, they must be treated as belonging to two diVerent

classes. Thus, as in the case of gender in Latin (remember nauta bonus in

section 5.2), classiWcation of nouns in these languages may have to be based

on agreement patterns rather than direct morphological properties.

The classiWcation of nouns in these languages may have some semantic

correlates.Class I includesmostpersonalnouns, class IInamesof trees, plants,

andanumberof inanimate entities.The singularpreWxof class IV isoftenused

for the names of languages. Thus, Swahili is also referred to as Ki-Swahili.

The noun-class system of Bantu languages is very important for agree-

ment patterns. A nice example of this pervasive agreement is the following

sentence from Swahili (Welmers 1973: 171):

(9) ki-kapu ki-kubwa ki-moja ki-lianguka

basket large one fell

‘‘One large basket fell’’

However, this kind of alliterative concord in Swahili is exceptional: it is not

necessarily the case that the same preWx appears on all agreeing constitu-

ents of a clause. In fact, it is only the class preWx for attributive modiWers

that always has the same preWx as the head noun. In other cases there are

diVerent preWxes involved as illustrated here for the class I noun m-tu

‘‘person’’ (Welmers 1973: 171):

(10) attributive m-tu m-moja ‘‘one person’’

associative m-tu w-a Utete ‘‘a person from Utete’’

demonstrative m-tu yu-le ‘‘that person’’

subject m-tu a-likuja ‘‘a person came’’

The African noun-class system certainly has a pretty complicated

appearance. Yet, just like gender, it is deeply rooted in the language system

and the language user. Loan words are incorporated into that system

(sometimes accompanied by reinterpretation of the phonological form of

the loan). For instance, the English instruction keep left has been adapted in


Swahili as kipilefti, and is used to denote a roundabout. The Wrst syllable ki

has been interpreted as a class preWx (class IV), and hence the plural form

of this word is vipilefti. A similar case is the Arabic loan word kitabi ‘‘book’’

with the plural form vitabi.

The other important inXectional category for nouns is case. Like number

properties, case properties of nouns are transferred to words of other

classes through agreement. The Icelandic paradigm in Table 6.1 is quite

characteristic of Indo-European case systems with respect to the number of

cases. Languages may diVer considerably in the number of cases they have,

if any. Finnish and Hungarian are well known for their large number

of cases. There are four cases with a clearly grammatical function:

nominative, accusative, genitive, and partitive. The partitive case is used

to mark patient objects that are partially aVected by the action, as the

following sentence from Hungarian illustrates (Blake 1994: 153):

(11) Olvasott a konyv-bol

read.3sg the book-part

‘‘He read some of the book’’

This language and its relatives are well known for its large set of local cases.

The Finnish nouns in (12) all have a particular local case. These local cases

have similar functions as locative prepositions in languages with less elab-

orated case systems such as Latin. In addition to the local cases, there are

also cases such as the abessive with the meaning ‘‘without’’, and the

comitative, with the meaning ‘‘accompanied by’’:

(12) Case label Meaning Example

allative to(wards) (the exterior) of poyda-lle ‘‘onto the table’’

illative into laitokse-en ‘‘to the institute’’

ablative from (the exterior of) kadu-lta ‘‘out of the street’’

elative from (the inside of) kaupa-sta ‘‘out of the shop’’

inessive in(side) talo-ssa ‘‘in the house’’

adessive at roof-ade ‘‘on the roof ’’

The elative and the partitive case are both used in quantifying expres-

sions. Consider the following examples from Finnish (Sulkala and Karja-

lainen 1992: 234):

(13) kaksi poji-sta kaksi poika-a

two boy.pl-elative two boy-partitive

‘‘two of the boys’’ ‘‘two boys’’

132 inflection

Note that in the second example, the noun is not marked for plural. The

number word functions as the head of the phrase, and assigns partitive case

to the following noun. These examples also serve to illustrate the relation

between particular case labels and their semantic interpretation.

The case labels here are primarily formal notions. A particular case

may be used for a number of diVerent meanings. The Latin ablative

case, for example, may be used with an instrumental meaning, but also to

indicate removal from somewhere. In Finnish, the instrumental meaning

can be expressed by the adessive case, as in (Sulkala and Karjalainen

1992: 224):

(14) Tein leiva-n konee-lla

make.impf.1sg bread-acc machine-adessive

‘‘I made bread with the machine’’

The genitive case is well known for its vague meaning: it may specify a

relation between the noun marked with genitive case and another noun (the

head noun), but the speciWc interpretation of that relation is a matter of

interpretation. The Latin phrase amor parentum ‘‘the parents’ love’’, with

the word parens ‘‘parent’’ in the gen.pl form, can either denote the love

that parents have (for their children), or the love that they receive from

their children. In the Wrst case, they are the subject of love, in the second

case the object. Hence, one Wnds the corresponding semantic distinctions

genitivus subjectivus and genitivus objectivus in traditional descriptions

of Latin. The general meaning of the genitive is clearly ‘‘relation’’. The

linguist Roman Jakobson has tried to derive such speciWc semantic inter-

pretations of each of the Russian cases from a general, vague meaning,

referred to as the Gesamtbedeutung (Jakobson 1936).

6.2 Categories of verbal inflection

There are three important categories of inherent inXection for verbs: tense,

mood, and aspect.Many languages have overt marking for these categories,

and in language descriptions one usually Wnds a description of the Tense–

Mood–Aspect system or TMA system for short. In addition, there is a

category Voice (such as Active versus Passive forms) that is sometimes

considered as part of verbal inXection. However, since this category has

eVects on the syntactic valency of verbs, it is treated in Chapter 8. Verbs


may also carry features of the participants of the event they denote. Hence,

we also Wnd number, person, and gender marking on verbs.

The tense of a verb locates the situation denoted by the relevant clause on

a time axis with, in most cases, the moment of speaking as the point of

reference. The notion tense is to be distinguished from the notion ‘time’.

With tense we refer to the grammatical expression of time notions in a

language, through morphology or periphrastic forms. Time, on the other

hand, can be expressed through all kinds of phrases, such as the previous

day. The past tense indicates that the situation obtained before the moment

of speaking, the present tense indicates that the situation obtains at the

moment of speaking, and the future tense signals that the situation denoted

is located on the time axis after the moment of speaking. Some languages

have speciWc morphological forms for these three tenses, as is the case for

Lithuanian (Chung and Timberlake 1985: 204):

(15) a. dirb-au

work-1sg.past

‘‘I worked/was working’’

b. dirb-u

work-1sg.pres

‘‘I work/am working’’

c. dirb-s-iu

work-fut-1sg

‘‘I will work/will be working’’

The function of tense is a deictic one, because its interpretation depends on

some external point of reference in the speech situation, the time axis. With

the notion deixis we refer to the link between elements of an utterance and

entities in some extra-linguistic reality. In the sentenceHe is ill, for instance,

the personal pronoun he is used deictically: the speaker assumes that the

hearer is able to identify the referent of that pronoun in the speech situation.

The interpretation of tense forms is sometimes more complicated than

sketched above. A well-known example of a mismatch between form and

interpretation is the use of the present tense in referring to the past. This use

is traditionally called praesens historicum. In this kind of language use, we

use present tense forms in speaking about the past, in order to make our

report on what has happened more lively, and incisive.

Another salient feature of tense is that, although it is marked on the verb,

its semantic scope is the whole clause to which it belongs. In the sentence

134 inflection

Suzanne told the children a fairy-tale, the whole situation of Suzanne telling

the children a fairy-tale is claimed to have been true at some moment before

the moment of speaking. The statement John met the Tsar can therefore

be true now, at the moment of speaking, even though there is no Tsar any

more.

The notion aspect refers to the way in which situations (states or events)

can be presented. The term is used as a semantic notion, but also to refer to

the grammatical expression of that semantic notion. As Comrie (1976: 3)

stated, ‘aspects are diVerent ways of viewing the internal temporal con-

stituency of a situation’. Perfective aspect presents a situation as completed,

whereas imperfective aspect presents the situation as ongoing. This kind of

aspect is referred to as grammatical aspect.

Grammatical aspect has to be distinguished from predicational aspect,

the kind of aspect that is determined by a verb and its arguments. Consider

the following English sentences:

(16) a. John reads a lot

b. John reads the book

The action of reading does not necessarily have an inherent endpoint, and

hence the verb to read in its intransitive use in (16a) receives a durative (or

atelic) interpretation. This durative aspect is the inherent lexical aspect of

this verb. In (16b), the same verb is used as a transitive verb, with a deWnite

direct object, and hence this sentence receives a telic interpretation, in

which the endpoint of the action is implied. Predicational aspect thus not

only depends on the nature of the verb, but also on that of its arguments

such as the object NP, and is therefore compositional in nature

For other verbs, telic aspect is an inherent property, as is the case for the

verb to die that denotes a punctual event. Hence, we call it lexical aspect

(also referred to by its German label, Aktionsart). A punctual event by

deWnition implies that the endpoint of the event is reached, and hence its

lexical aspect is telic. Verbs with inherent telic aspect can also be created by

means of derivational morphology, as we will see in Chapter 9.

In sum, there are at least three diVerent layers involved in the aspectual

properties of a sentence: (i) lexical aspect, (ii) predicational aspect, and

(iii) grammatical aspect, the latter being a property of the proposition

expressed by the clause.

The necessity of distinguishing between grammatical aspect and predica-

tional aspect is illustrated by the following sentence:


(17) John is reading the book

The predicate reading the book has telic aspect, since there is an implied

endpoint for the action denoted. Yet, its grammatical aspect is imperfect-

ive: it presents the situation as ongoing, marked by the periphrastic

progressive form of the English verb.

The notion perfect must be distinguished from the notion perfective

aspect. The perfect relates two diVerent points on the time axis, and

‘indicates the continuing relevance of a past situation’ (Comrie 1976: 52).

Therefore, the perfect is often considered a temporal rather than an aspect-

ual distinction. Compare the following two English sentences:

(18) a. past tense: Herry read the last novel by McEwan

b. present perfect tense: Herry has read the last novel by McEwan

The Wrst sentence denotes an individual event. In the second sentence, it is

indicated that the event in the past of Herry reading a particular novel

is added to the inventory of achievements by Herry. This explains why a

temporal adverbial such as yesterday can be used in combination with

the past tense, but not in combination with the present perfect: a time

speciWcation is only correct for an individual event.

In some European languages the diVerences between past and present

perfect have become blurred. For instance, in Dutch, unlike English, it is

possible to use the present perfect form in combination with gisteren

‘‘yesterday’’, as in:

(19) Gisteren heeft Herry de laatste roman van McEwan gelezen

Yesterday has Herry the last novel by McEwan read

‘‘Yesterday, Herry read the last novel by McEwan’’

In Afrikaans, the simple past tense forms of its mother language Dutch

have disappeared completely and the periphrastic present perfect forms are

used to express the simple past.

The distinction between imperfective and perfective aspect is most com-

mon for past tense forms; present tense denotes an ongoing event, and hence

is imperfective by nature. The following examples from Spanish illustrate the

diVerences in interpretation between the two aspects (Gonzalez 2003: 112):

(20) a. Laura beb-io una Coca-Cola

Laura drink-past.perf a Coca-Cola

‘‘Laura drank a Coca-Cola’’

136 inflection

b. Laura beb-ıa una Coca-Cola

Laura drink-past.impf a Coca-Cola

‘‘Laura drank a Coca-Cola’’

In (20a), the situation is presented as closed oV, whereas in (20b) the event

is interpreted as ongoing, with a progressive or habitual interpretation.

These examples also illustrate that, quite often, the diVerent forms of a

verbal paradigm express both tense and aspect by means of the same

ending, that is, by means of cumulative exponence.

The third category associated with verbal paradigms is that of mood.

Mood describes the actuality of an event. For instance, a language may use

morphology to distinguish between actual and non-actual events by having

a realis and an irrealismood. The indicativemood is typically the mood for

realis, whereas subjunctive and imperative forms denote some sort of non-

actuality. Subjunctive forms are found, for instance, in embedded clauses

with a matrix verb of hoping, ordering, or claiming. The following example

illustrates this for German:

(21) Er behauptet er sei dort gewesen

he claims he be.subjunctive there been

‘‘He claims he has been there’’

If you hope or think that something will happen, or if you order something

to happen, it has not yet happened, so the event is not part of the actual

world. Mood may also be used to indicate the epistemic status of the

proposition. Latvian, for instance, has an evidentialmood to report hearsay

knowledge, as in the following sentence (Nau 2001: 189):

(22) Cit-i saka tu es-ot visai skarb-s

other-pl say.pres.3 2sg be-evid very harsh-nom.sg.masc

tip-s

fellow-nom.sg

‘‘Some say—you are a rather harsh fellow’’

Although tense, aspect, and mood are qualitatively diVerent notions, they

are often intertwined in the form of a verb. For instance, past tense and

perfective aspect are related, because for an event to be completed, it

cannot be ongoing, and thus excludes present tense. There is also a relation

between past and irrealis: the event denoted is not an actual one. Hence,

past tense forms are used to express irrealis, as in:

(23) I would have come, if I had been able to


In combination with the future modal auxiliary, a counterfactual

interpretation is invoked.

Because these three notions are so intertwined, one often Wnds the term

TMA-system to refer to the verbal system for expressing these three cat-

egories and related ones such as speaker’s attitude. In so far these three

categories are expressed by diVerent morphemes, the following tendency

has been observed concerning their ordering and that of person markers

with respect to the verbal stem:

(24) aspect > tense > mood > person (Bybee 1985: 35)

This hierarchy means that aspect markers tend to be found more closely to

the stem than tense markers, tense markers more closely than mood mark-

ers, etc. As argued by Bybee (1985), this ordering may reXect the degree to

which the meaning expressed by the verbal root is aVected by these diVerent

markers. For instance, aspectual markers, which are closest to the verbal

root, have a direct eVect on the kind of event that is denoted by the verb.

Tense markers, on the other hand, do not aVect the meaning of the verb

directly, but express the location of the state or event denoted by the verb

on the time axis with respect to the moment of speaking.

In addition to these three categories of inherent inXection on verbs, verbs

may exhibit contextual inXection in the form of person, number, and

gender marking. The distinction between three persons for both singular

and plural number is quite pervasive in the languages of the world. Person

marking is found for the subject, and sometimes also for other dependents

of the verb such as the object. The number of distinctions made by a

particular language can sometimes be read oV its system of pronouns.

The next step is to see how, and to what extent, these distinctions are

also encoded in verbal forms. Many pronominal systems have up to six

diVerent forms for these six possible combinations of person and number.

The semantic interpretation of the distinction between the three persons is

as follows:

(25) 1 ¼ speaker, 2 ¼ addressee, 3 ¼ any other entity than speaker or addressee

These semantic qualiWcationsmake clear that person is a deictic notion. The

referential value of personal pronouns, for instance, is not Wxed, and de-

pends on the situation in which the sentence is uttered. That is why the

Danish linguist Jespersen called these pronouns shifters. The referential

value of, for instance, I will shift continuously in the course of a dialogue.

138 inflection

For a child, it sometimes takes some time before it has mastered that

system, and understands that I can refer to someone else than the child itself.

The notion plural has a particular interpretation in connection to person,

namely that of ‘‘group’’. The following combinations are logically possible

(after Cysouw 2001: 70):

(26) 1 + 1 ‘‘we’’, mass speaking

1 + 2 ‘‘we’’, including addressee, excluding other

1 + 3 ‘‘we’’, including other, excluding addressee

1 + 2 + 3 ‘‘we’’, complete

2 + 2 ‘‘you-all’’, only present audience

2 + 3 ‘‘you-all’’, addressee(s) and others

3 + 3 ‘‘they’’

The English pronoun we allows for all four logically possible interpret-

ations; the Wrst one is only appropriate when people speak or sing together

as in the soccer stadium where the audience sings We are the champions.

Some languages have diVerent forms for inclusive we (1 + 2) versus exclu-

sive we (1 + 3), as illustrated here for the possessive pronouns of Cuzco

Quechua (van de Kerke 1996: 121):

(27) wasi-y ‘‘my house’’

wasi-yki ‘‘your house’’

wasi-n ‘‘his house’’

wasi-nchis ‘‘our (speaker’s and hearer’s) house’’

wasi-yku ‘‘our (speaker’s and not hearer’s) house’’

wasi-ykichis ‘‘your house’’

wasi-nku ‘‘their house’’

Many languages that have six formal distinctions in their pronominal

system do not have the same number in their verbal inXection. In the

English singular present tense forms of regular verbs we only Wnd a formal

morphological distinction between third person versus Wrst/second person

(walk-s versus walk); in the plural, there are no morphologically marked

person distinctions whatsoever. This kind of syncretism can be accounted

for by considering such forms as underspeciWed for certain morphosyntac-

tic features. The English plural forms are not speciWed for the category

person, and hence they combine with subjects of all three persons. When

combinations of Wnite verbal forms and subjects are checked by the rule of

Agreement, these plural forms will be discovered to be compatible with

their subject with respect to these categories. The form walk can be speciWed


as [-3sg], and will therefore be combinable with 1sg and 2sg, and with all

kinds of plural subjects. Italian, on the other hand, has distinct present tense

forms for all six person–number combinations.

Some linguists have claimed that there is a relation between the omissa-

bility of the subject and the richness of the verbal paradigm. Italian, for

instance, has six diVerent present tense forms, and allows for omission of the

subject. This idea is referred to as the Pro-drop-parameter: languages diVer

as to the obligatory expression of the subject, depending on the degree of

richness of the verbal paradigm. However, it appears that there is no

necessary connection between the two since there are also languages that

do not have diVerent forms for each of the six cells, and yet do allow for

omission of the subject, as is the case for Indonesian (Cysouw 2001: 52).

In many languages, politeness conventions aVect the selection of a

particular person form. In Italian, for instance, the 3sg form is used as

the form for addressing persons in a polite way:

(28) Come sta-i? vs Come sta?

how be-2sg how be.3sg

‘‘How are you?’’ ‘‘How are you?’’ (polite)

The 3sg form creates a larger distance between speaker and addressee than

a 2sg form. This distance or indirectness receives the value of politeness.

Other languages use plural forms for that purpose, or a combination of

both, as in German where both the 3pl subject pronoun and the corre-

sponding verbal form are used to refer to a singular or plural addressee in

order to convey politeness:

(29) sie kommen ‘‘they come’’ Sie kommen ‘‘you (sg/pl) come’’ (polite)

du komm-st ‘‘you (sg) come’’ (informal)

The use of a capital initial for the pronoun in the politeness form is an

orthographic marker of this use of the pronoun. In old-fashioned Dutch,

the 1pl verbal form in combination with the pronounwij ‘‘we’’ can be used as

a pluralis majestatis, as in the opening line of Dutch law texts which are

formulated as a decision of the Dutch queen:Wij Beatrix, bij de gratie Gods

Koningin der Nederlanden ‘‘We, Beatrix, by the grace of God Queen of

the Netherlands’’. The 1pl form can also be used by writers to express

modesty, since written Ik ‘‘I’’ might be felt as the author imposing him/herself

too much on the reader. Thus, the use and interpretation of person–number

features is governed by language-speciWc conventional pragmatic rules.

140 inflection

6.3 Autonomous morphology

In the realm of inXection, the relationships between form and meaning are

quite complex, with much formal variation that does not serve to directly

express a speciWc grammatical content. The existence of declensions and

conjugations means that there is a many-to-one relationship between form

and meaning. We might refer to these phenomena as ‘autonomous morph-

ology’ since morphology has its own formal systematics that does not bear

upon the other modules of the grammar. Let us have a closer look at these

complexities in the domain of verbal inXection.

Many Indo-European languages have a number of verbal conjugations.

Italian, for instance, has three conjugations, each with a diVerent thematic

vowel after the root, a, e, or i. The third conjugation has two subclasses

because some of the i-verbs have a special allomorph with an augment isc.

We call this string isc an augment since it does not express some inXectional

property. The verbal forms in Table 6.3 constitute only a subset of the

diVerent verbal paradigms, and are meant as a modest illustration of

the possible richness of verbal inXectional systems. The occurrence of the

augment isc is lexically governed but not completely arbitrary. It only

appears in those forms in which otherwise the stress would fall on the

root (sg and 3pl forms).

In Italian, the inWnitive form is the form used to refer to a particular

verb. This form contains the thematic vowel that we have to know in order

to compute the whole set of forms in a paradigm. So the inWnitive form has

the structure ‘root + thematic vowel + ending’, and it is the stem (root +

thematic vowel) that is the basis for computing the diVerent forms. These

diVerent thematic vowels show up in a systematic way in the imperfective

past tense forms, which are completely regular once you know the thematic

vowel.

In order to compute the whole verbal paradigm, you may actually need

more than one stem-form for the computation of the whole verbal para-

digm. Consider the Latin data in Table 6.4. As these data show, the best

way to compute the future participle is deriving it from the perfect participle

by adding the suYx -ur. In those cases in which the perfect participle has an

irregular form, that irregularity recurs in the future participle. The clearest

case of this argument is the verb ferre that has a suppletive stem for the

perfect participle. This suppletive stem lat- recurs in the future participle.


However, from a semantic point of view, there is no ground for claiming

that the future participle is derived from the perfect participle. Obviously,

the future participle does not carry a perfect meaning. So we have to say

instead that Latin verbs have (at least) two stems. Stem-1 is the stem-form

on which the inWnitive is based. Stem-2 forms the basis for both the perfect

participle and the future participle. This is represented in Figure 6.1. For

regular verbs such as laudare, Stem-2 can be computed from Stem-1

(through the addition of -t). For irregular verbs, Stem-2 has to be stored

Table 6.3. The Italian verb paradigm

infinitive

parlare

‘‘speak’’

temere

‘‘fear’’

dormire

‘‘sleep’’

capire

‘‘understand’’

present

1sg parlo temo dormo capisco

2sg parli temi dormi capisci

3sg parla teme dorme capisce

1pl parliamo temiamo dormiamo capiamo

2pl parlate temete dormite capite

3pl parlano temono dormono capiscono

past

imperfective

1sg parlavo temevo dormivo capivo

2sg parlavi temevi dormivi capivi

3sg parlava temeva dormiva capiva

1pl parlavamo temevamo dormivamo capivamo

2pl parlavate temevate dormivate capivate

3pl parlavano temevano dormivano capivano

past perfective

1sg parlai temei dormii capii

2sg parlasti temesti dormisti capisti

3sg parlo teme dormı capı

1pl parlammo tememmo dormimmo capimmo

2pl parlaste temeste dormiste capiste

3pl parlarono temerono dormirono capirono

future

1sg parlero temero dormiro capiro

2sg parlerai temerai dormirai capirai

3sg parlera temera dormira capira

1pl parleremo temeremo dormiremo capiremo

2pl parlerete temerete dormirete capirete

3pl parleranno temeranno dormiranno capiranno

142 inflection

as part of the lexical information on that verb. AronoV (1994) introduced

the term morphome as a label for categories such as Stem-2 which have no

grammatical relevance beyond the functioning of morphology. Stem-2 is a

morphome, a morphological function that speciWes the stem form for

particular morphosyntactic forms.

DiVerent verbal stems also play a role in word-formation. For instance,

the stem of Italian perfect participles is also used for coining deverbal agent

nouns (Vogel 1994: 233), see (30):

(30) Verb Past participle Agent noun

corregg-ere ‘‘to correct’’ corrett-o corrett-ore

dirig-ere ‘‘to direct’’ dirett-o dirett-ore

distrugg-ere ‘‘to destroy’’ distrutt-o distrutt-ore

scriv-ere ‘‘to write’’ scritt-o scritt-ore

Another kind of formal variation in paradigms is that of suppletion.

Suppletion underscores the importance of the notion ‘inXectional para-

digm’: forms belong together, not always because they are phonologically

similar, but because they Wll the cells of the paradigm of the same lexeme.

Classic cases of suppletion are the diVerent forms of the verb to be in

English. In many languages, the equivalent of this verb has a suppletive

Table 6.4. Latin verbal stem-forms

infinitive perfect participle future participle gloss

lauda-re laudat- laudat-ur- praise

mone-re monit- monit-ur- warn

duce-re duct- duct-ur- lead

prem-ere press- press-ur- press

fer-re lat- lat-ur- carry

Source: AronoV 1994: 32.

Fig. 6.1 Stems in Latin


paradigm (Table 6.5). Suppletive forms usually occur in particular with

lexemes with a high frequency of use. In a number of languages, the verb

for ‘‘to give’’ has diVerent stems depending on the person of the recipient:

Wrst and second person versus third person. In Kolyma Yukaghir, a lan-

guage of northeastern Siberia, the following pair of suppletive forms is

found (Elena Maslova, cited in Comrie 2001):

(31) tadi- ‘‘give to 3pers’’, kej- ‘‘give to 1/2pers’’

In languages with a number of conjugations, there is usually one default

conjugation: the inXection class taken by new verbs that enter the language.

In Italian, this is the a-conjugation. A somewhat diVerent situation obtains

in Germanic languages. They have a default conjugation in which the past

tense stem is formed through suYxation. This is called the weak conjuga-

tion, illustrated in Table 6.6.

In addition they have a number of classes of strong or stem-alternating

verbs that create past tense by means of vowel alternation (Ablaut). The

past participles of weak and strong verbs also diVer in that those of strong

verbs exhibit vowel alternations as well (Table 6.7). Weak verbs have a

past tense suYx -te after stems ending in a voiceless obstruent, and -de

elsewhere. The past participle is formed by preWxing the stem with ge-, and

suYxing it with -t (again, after voiceless obstruents), and -d elsewhere. Note

that the past participles of stem-alternating verbs have another suYx, -en.

Table 6.5. Singular present forms of ‘‘to be’’

English Dutch French Italian

1sg am ben suis sono

2sg are bent es sei

3sg is is est e

Table 6.6. Regular past tense and participle formation in Dutch

verb stem past tense sg past participle

tob ‘‘to toil’’ /tOb/ tob-de [bd] ge-tob-d

leg ‘‘to lay’’ /lEg/ leg-de [gd] ge-leg-d

kap ‘‘to cut’’ /kAp/ kap-te [pt] ge-kap-t

paf ‘‘to puV ’’ /pAf/ paf-te [ft] ge-paf-t

144 inflection

New verbs that enter the language (through borrowing or conversion from

nouns) are always inXected as weak verbs. This is independent evidence for

the weak verbs—which form the largest inXectional class of all—being the

default class. For instance, the Dutch verb computer ‘‘to use a computer’’

has computer-de as its past tense stem. The verb prijzen ‘‘to praise’’ is a

strong verb. There is also a homophonous verb prijzen derived through

conversion form the noun prijs ‘‘price’’, with the meaning ‘‘to price, to

provide with price labels’’. This latter verb is weakly inXected, and thus we

get the pair of past tense singular forms prees [pre:s] ‘‘praised’’ versus

prijsde [prEizd

e

] ‘‘priced’’.

As in the case of suppletion, the verbs that exhibit apophony tend to

belong to the verbs with the highest frequency of use. The entrenchment of

these past tense stems in the memory of the language user must be an

important factor for the relative stability of such complicated systems.

Finally, periphrasis is a common feature of verbal inXectional systems.

For instance, in many European languages forms of the verb for ‘‘to have’’

and ‘‘to be’’ are used as auxiliaries in periphrastic forms for expressing the

Table 6.7. Stem alternating verbs in Dutch

Alternation pattern Present tense stem Past tense stem Past participle

A-B-B

/Ei-e:-e:/ knijp ‘‘to pinch’’ kneep ge-knep-en

/i-o:-o:/ schiet ‘‘to shoot’’ schoot ge-schot-en

/Vy-o:-o:/ buig ‘‘to bend’’ boog ge-bog-en

/e:-o:-o:/ weeg ‘‘to weigh’’ woog ge-wog-en

/I-O-O/ bind ‘‘to bind’’ bond ge-bond-en

/E-O-O/ zend ‘‘to send’’ zond ge-zond-en

A-B-A

/a:-u-a:/ draag ‘‘to carry’’ droeg ge-drag-en

/a:-i-a:/ blaas ‘‘to blow’’ blies ge-blaz-en

/A-i-A/ val ‘‘to fall’’ viel ge-vall-en

/A-I-A/ hang ‘‘to hang’’ hing ge-hang-en

/o:-i-o:/ loop ‘‘to walk’’ liep ge-lop-en

/O-E-O/ word ‘‘to become’’ werd ge-word-en

/u-i-u/ roep ‘‘to call’’ riep ge-roep-en

A-B-C

/E-i-O/ help ‘‘to help’’ hielp ge-holp-en

/E-i-a:/ schep ‘‘to create’’ schiep ge-schap-en

/e:-u-o:/ zweer ‘‘to swear’’ zwoer ge-zwor-en


perfect, perfective aspect, or past tense. Historically, the use of to have as an

auxiliary derives form its use in predicative constructions. The sentence

I have the door shut can be interpreted as parallel to I keep the door shut,

with shut specifying a property of the door, something like ‘‘I have the door

in a shut state’’. Such constructions could be reanalysed as expressing that

a certain state of the object has been achieved, thus providing an interpret-

ation of ‘‘completion of the act of shutting’’. The combination of auxiliary

and past participle thus started leading a life of its own.

Summary

Systems for the inXections of nouns, adjectives, and verbs show compli-

cated patterns of formal variation in the expression of morphosyntactic

categories. That is why often a number of declensions and conjugations

have to be distinguished. The main inXectional categories for nouns and

adjectives are gender, number, and case. Verbs may be morphologically

marked for tense, aspect, mood, gender, person, and number.

The facts discussed in this chapter show that the notion ‘paradigm’ plays

an indispensable role in the analysis of inXectional systems and their sub-

systems. Notions such as periphrasis, suppletion, and rule of referral all

presuppose that inXectional forms are conceived of as Wlling the cells of a

paradigm.

The notion ‘autonomous morphology’ introduced in section 6.3 pertains

to the whole range of purely formal regularities involved in the proper

morphological expression of grammatical content on words, and the

analysis of these regularities.

Questions

1. The Animacy Hierarchy plays a role in the use of English personal pronouns.English distinguishes three genders for 3sg personal pronouns: he, she, itthat can be used to refer to NPs. However, this threefold gender distinctionis not maintained for all antecedent NPs of these pronouns. Where on thehierarchy is the cut-off point for the three-gender distinction?

2. Consider the following fourteen singular case forms of the Finnish wordskasi ‘‘hand’’ and tytto ‘‘girl’’ (Sulkala and Karjalainen 1992: 386):

nominative kasi tyttogenitive kaden tyton

146 inflection

partitive katta tyttoaessive katena tyttonatranslative kadeksi tytoksiabessive kadetta tytottainessive kadessa tytossaelative kadesta tytostaillative kateen tyttoonadessive kadella tytollaablative kadelta tytoltaallative kadelle tytolleinstrumental kaden —comitative kasineen tyttoineen

a. List the different stem shapes of these two words, and their distributionacross the cases.

b. If we wanted to make the instrumental case form of tytto ‘‘girl’’, whatwould it be?

3. Lithuanian has five different declensions for nouns. Nouns with a nom.sg

form in -as, -is, or -ys such as vyras ‘‘man’’, brolis ‘‘brother’’, and arklys‘‘horse’’ belong to the first declension, and have the paradigms shown inthe table below:

a. What are the stem-forms of these three nouns?

b. What rule(s) of referral can be formulated for these paradigms?

4. My Latin school grammar presents the following paradigms for the wordsnemo ‘‘nobody’’ and nihil ‘‘nothing’’:

nominative nemo nihilgenitive nullıus nullıus reıdative neminı nullı reıaccusative neminem nihilablative a nullo nulla re

Which paradigmatic notions are presupposed in this presentation of theforms of nemo and nihil?

singular plural

nom vyras brolis arklys vyrai broliai arkliaigen vyro brolio arklio vyru broliu arkliudat vyrui broliui arkliui vyrams broliams arkliamsacc vyra broli arkli vyrus brolius arkliusinst vyru broliu arkliu vyrais broliais arkliaisloc vyre brolyje arklyje vyruose broliuose arkliuosevoc vyre broli arkly vyrai broliai arkliai

Source: Tekoriene 1990: 221; the y stands for [i :], and the cedille on vowel letters indicateslength.


5. In some languages, the imperfect or past tense can be used with modal orpoliteness overtones, as in the following sentences fromDutch and Finnish(Sulkala Karjalainen 1992: 299) respectively:

Wat was uw naam?what was your name‘‘What is your name, please?’’Oliko rouvala lippua?be.IMPF.3SG.Q madam.ADESSIVE ticket.PARTITIVE‘‘Would you have a ticket, madam?’’

How would you explain this use of past tense forms?

6. Which form of systematic syncretismmay be assumed for the Polish nounsin (1) of Chapter 2?

7. In the following Latin sentences, the verbal infinitive used as subjectimposes neuter gender on the predicative adjective, and person andnumber marking on the verb:

mentı-ri turp-e estlie-INF scandalous-NEUT be.3SG

‘‘Lying is scandalous’’naviga-re difficil-e estsail-INF difficult-NEUT be.3SG

‘‘Sailing is difficult’’

Explain why these verbal infinitives trigger agreement, although it istypically nominal elements that are the sources of the agreeing features.

8. In Turkish constructions with partitive numerals, the quantified noun re-ceives genitive or ablative case. The numeral itself carries a nominal agree-ment marker (Kornfilt 1997: 236–7).

cocuk-lar-in iki-si cocuk-lar-dan iki-sichild-PL-GEN two-3SG child-PL-ABL two-3SG‘‘two of the children’’ ‘‘two of the children’’

Is the case marking of the word cocuk-lar ‘‘children’’ an instance of depen-dent marking or of head marking?

9. Compare the following Dutch sentence and its English gloss:

Morgen breng ik je het boektomorrow bring I you the book‘‘Tomorrow, I’ll bring you the book’’

Howmight it be explained that the future tense which is expressed overtlyin English does not receive overt marking in Dutch?

10. The following Turkish sentence has an iterative aspectual interpretationalthough there is no morphological marker for iterative aspect (Kornfilt1997: 360):

Hasan oksur-uyorHasan cough-PRES.PROGR

‘‘Hasan is coughing repeatedly’’

How can this iterative aspectual interpretation be explained?

148 inflection

Further reading

There are monographs on the following morphosyntactic categories discussedin this chapter: Number (Corbett 2000), Gender (Corbett 1991), Case (Blake1994), Tense (Comrie 1984), Aspect (Comrie 1976), Person (Cysouw 2001).Chung and Timberlake (1985) is an article with a typological survey of TMA-systems. The role of gender in natural language is discussed in Unterbeck et al.(2000).For rules of referral, see Stump (1993, 2001). In a number of publications,

Carstairs-McCarthy has studied the constraints on the extent of allomorphy wefind in inflectional systems (for instance, Carstairs 1987; Carstairs-McCarthy1994; Cameron-Faulkner and Carstairs-McCarthy 2000).A formal analysis of syncretism by means of inheritance trees is proposed in

Corbett and Fraser (1993). Harley and Noyer (2003) discuss syncretism in theframework of Distributed Morphology. A website with a database on syncre-tism is http://www.smg.surrey.ac.uk.


http://www.smg.surrey.ac.uk


Part IVInterfaces


7The interface between

morphology and phonology

7.1 Morphology and phonology

The English adjective selective can be suYxed with either -ity or -ness,

resulting in selectıvity and selectiveness respectively. The acute accents on

these words indicate the location of main stress. As you can see, the attach-

ment of the suYx -ity has the eVect that the location of the main word

stress shifts rightwards, to the last syllable of the stem selective, whereas the

attachment of the suYx -ness does not aVect the location of the main stress

on the stem. This suggests that morphological structure may play a role in

determining the phonological form of a complex word. In this chapter we

will zoom in on the issue how morphological structure plays a role in com-

puting the phonological form of a word. Inversely, phonological properties

of words may also play a role in selecting an aYx with which it can com-

bine. The English suYx -al, for example, can only be attached to verbs that

end in a stressed syllable (arrıve–arrival, recıte–recital, chatter–*chatter-al ).

These kinds of interaction between morphology and phonology show that

there must be an interface between the morphological and the phonological

7.1 Morphology and phonology 153

7.2 Interface principles 156

7.3 Allomorphy and aYx

competition 168

7.4 Cyclicity and

co-phonologies 175

7.5 The morphological use

of phonology 177

Summary 182

Questions 182

Further reading 184

properties of words. ‘Interface’ means that diVerent kinds of information

about linguistic constructs (in these examples words) can ‘see’ each other.

In order to provide some more substance to the notion ‘interface’ in the

domain of morphology, we will Wrst consider what kinds of information on

words the grammar needs to provide. Aword is a complex piece of informa-

tion. It links a particular sequence of sounds to a particular meaning, and

also has formal properties such as a syntactic category label. The informa-

tion contained in the English simplex word dog, for instance, can be repre-

sented as in Figure 7.1. The Wrst piece of information in Figure 7.1 concerns

the phonological properties of this word: it is a phonological word (ø) that

consists of one syllable (�) that in its turn consists of a sequence of three

sounds. This phonological word bears the same index as the syntactic

information about this word (that it is a noun), and the semantic informa-

tion that it expresses the predicate DOG. Coindexation is used here to

specify the correspondence between the three kinds of information

involved in knowing a word.We thus see that a word has a tripartite parallel

structure.

Let us now look at a complex word such as the English word baker, a

noun derived from the verb bake through suYxation with -er. The three

kinds of information (the phonological form, the morphological structure,

and the meaning) concerning this word can be represented as in Figure 7.2.

The phonological structure of baker is that of a phonological word consist-

ing of two syllables, (be:)� and (k

e

r)�, and of Wve phonological segments. Its

Fig. 7.1 The representation of dog

Fig. 7.2 The representation of baker

154 interfaces

formal structure is that of a deverbal noun, as indicated by the tree that

represents its formal morphological structure.

The representation in Figure 7.2 may be generalized into a template for

nouns derived from verbs by means of the suYx -er. This is achieved by

omitting the word-speciWc information. This morphological template thus

speciWes that there is the following systematic relation between the three

kinds of linguistic information involved (Figure 7.3). In Figure 7.3 the level

of the syllables has been omitted because the number of syllables of words

ending in -er is not Wxed, but depends on the phonological make-up of

the base verb. The syllabiWcation of English words is predictable, and

need not be speciWed in morphological templates. Hence it is a computable,

predictable property of each individual deverbal noun in -er. Instead of the

speciWc predicate BAKE, the general label V is used to refer to the semantic

properties of the base verb.

The tripartite structure in Figure 7.3, an instance of a word-formation

template, is meant to make clear that morphology is not a module of

grammar on a par with the phonological or the syntactic module, which

are modules that deal with one aspect of linguistic structure only. Morph-

ology is word grammar, and similar to sentence grammar in its dealing with

the relationships between three kinds of information. It is only with respect

to the domain of linguistic entities that morphology is diVerent from

sentence grammar: morphology has the word domain as its primary focus.

This short introduction to the idea of tripartite parallel structure paves the

way for grasping the notion ‘interface’. This notion refers to theways inwhich

properties of one kind of structure relate to those of another structure. An

example of a relation between phonological and morphological form is that

the suYx -er is one of the so-called cohering suYxes of English. This means

that this suYx forms one domain of syllabiWcation with the stem to which it

has been attached. The word baker is syllabiWed in the same way as the word

father in which the sequence -er is not a suYx. The sound sequence -er forms

one syllable with the preceding consonant in both words: ba.ker, fa.ther

(remember that dots indicate syllable boundaries). Thus, the morphological

Fig. 7.3 The template for deverbal -er

the interface between morphology and phonology 155

boundary between bak- and -er in baker is not respected in phonology,

in the sense that it does not coincide with a syllable boundary.

There are also aYxes that do inXuence the way that a complex word is

syllabiWed. The English suYx -less, for example, is a non-cohering suYx.

This means that this suYx forms its own domain of syllabiWcation. The

adjective help-less, for instance, is syllabiWed as help.less, with a syllable

boundary coinciding with the internal morphological boundary. Compare

the syllabiWcation of this adjective to the syllabiWcation of the word duplex,

which is du.plex, with a syllable boundary before the consonant cluster /pl/.

The distinction between cohering aYxes and non-cohering ones is therefore

a theoretical distinction that we need for a proper account of the interface

between morphology and phonology.

These introductory remarks should give you some idea of what is meant

by ‘interface’. In this and the next two chapters, these interface issues are

dealt with in more detail.

7.2 Interface principles

An important task of the phonological module of a grammar is computing

the phonetic form of complex words. Consider the examples in (1) of plural

noun formation in Dutch. The plural nouns are formed by adding the suYx

-en/

e

n/ to the stem of the noun; the singular form has no overt phonological

marking. The basic procedure for computing the phonetic forms of these

plural nouns consists of three steps. The Wrst step is attaching the string of

segments of the plural suYx to the stem. This is a morphological operation.

The next two steps are phonological operations. Step 2 is the computation

of the prosodic structure of a word, in particular the way in which a word is

syllabiWed. In step 3, we scan the singular and plural forms as to the applic-

ability of phonological rules or constraints. A well-known phonological

(1) singular Phonetic form plural Phonetic form

hoed ‘‘hat’’ [hut] hoed-en [hud

e

n]

voet ‘‘foot’’ [vut] voet-en [vut

e

n]

poes ‘‘cat’’ [pus] poez-en [puz

e

n]

spies ‘‘spear’’ [spis] spies-en [spis

e

n]

156 interfaces

constraint of Dutch and German is that obstruents (stops and fricatives)

are voiceless at the end of a syllable. Therefore, the Wnal obstruents in the

singular forms of these nouns must be voiceless. In the plural nouns hoeden

and poezen, on the other hand, the stem-Wnal voiced obstruents appear at

the beginning of the second syllable, and hence they are not subject to

devoicing. The three steps are illustrated here for the singular and plural

forms of the word hoed ‘‘hat’’ (� ¼ syllable):

(2) step 1: morphology hud hud-

e

n

step 2: syllabiWcation (hud)� (hu)�(d

e

n)�

step 3: syllable-Wnal devoicing (hut)� not applicable

In step 1 we make use of the underlying form of the word hoed, the abstract

phonological form from which the diVerent surface forms of this word can

be derived. At the end of the derivation we have computed the phonetic

form of a word. We thus see that Dutch noun stems may exhibit allomor-

phy, variation in their phonological shape. The lexical morpheme/hud/ has

two diVerent shapes, [hut] and [hud]. This variation is governed by a

phonological constraint of Dutch, and hence this allomorphy is the pre-

dictable eVect of the phonological system of Dutch.

The plural form hoeden [hud

e

n] ‘‘hats’’ also serves to illustrate a general

point concerning the interface between phonology and morphology: the

potential asymmetry between morphological and phonological structure.

The word hoeden consists of Wve segments that are structured in two ways,

as shown in Figure 7.4. The representation of phonological structure in

Figure 7.4 requires some explication. The basic idea is that the sounds

of a word are organized into higher units. Sound segments combine into

syllables (�), syllables into feet (F), and feet into phonological words (ø).

The foot in this word is a trochee, that is, a foot consisting of two syllables

Fig. 7.4 The morphological and phonological structure of hoeden


of which the Wrst is the head and carries stress (hoeden carries stress on its

Wrst syllable). In this case, the phonological word happens to consist of only

one foot. This hierarchical organization of a word’s segments is also called

its prosodic structure, and instead of ‘phonological word’ the term prosodic

word may be used.

A basic constraint on the relation between lexical words (that is, non-

function words) and prosodic words is that a lexical word must consist of at

least one prosodic word:

(3) Lexical word ¼ minimally prosodic word

English and Dutch require a prosodic word to contain at least one full, that

is, non-reduced vowel. Hence, they cannot have schwa [

e

] as their only

vowel, unlike function words such as a and the. Dutch function words such

as een [

e

n] ‘‘a’’ and er [

e

r] ‘‘there’’ violate a second constraint of Dutch,

namely that a prosodic word cannot begin with a schwa. Hence, unlike

these function words, lexical words of Dutch never begin with a schwa.

The asymmetry of phonological and morphological structure manifests

itself quite clearly in Figure 7.4 with respect to the /d/: at the level of

morphological structure it forms a unit with the preceding sounds, at the

level of phonological structure it combines with the following sounds. The

interaction between morphology and phonology in this example is, so it

seems, zero. Phonology does not seem to care about the formal morpho-

logical structure of this word. However, as we will see below, there are many

cases in which morphological structure does inXuence the phonological

form of a word.

The three steps in (2) illustrate the idea of phonological derivation: the

computation of the phonetic forms of words in a number of steps, which

is a hallmark of classical generative phonology. There is an alternative,

non-derivational model that can be used to achieve the same result. In that

model, the phonology of a language is seen as a set of ranked constraints.

In the case of hoeden, three constraints are relevant. One is the constraint

which demands that obstruents are voiceless in syllable-Wnal position. Let

us refer to it as FinDevoicing. A second constraint is called Faithfulness: the

phonetic realization of a word or morpheme should be identical to its

underlying form, and not deviate from that underlying form. That is,

allomorphy should be avoided. A third constraint that plays a role is that

syllables should begin with a consonant. This is the No Empty Onset

constraint. As the phonetic form of the singular form hoed [hut] shows, in

158 interfaces

Dutch the constraint FinDevoicing is ranked higher than Faithfulness since

we do get allomorphy. The existence of allomorphy shows that constraints

can be violated: faithfulness is violated in the singular form in order to

satisfy the higher ranked constraint FinDevoicing. The selection of the

optimal phonetic form of hoed and hoeden is shown in Figure 7.5. It is

represented in tables (called tableaux), and this variety of phonological

analysis is called Optimality Theory.

Fig. 7.5 OT-tableaux for hoed and hoeden

The left columns mention the possible phonetic forms for the underlying

forms (given between slashes). These phonetic forms are called the candi-

dates. The left–right order of the constraints represents their ranking. The

leftmost is the highest ranked one. The asterisks in the cells of the tableaux

indicate that a constraint is violated by the candidate phonetic form. If a

constraint is no longer relevant for choosing the optimal candidate, the

corresponding cell is shaded. The pointed Wnger indicates the optimal

phonetic form. In the case of hoed the second candidate is selected since

the Wrst candidate violates a higher ranked constraint than the second one.

The exclamation mark indicates that a violation is fatal. That is, this

violation results in the fact that the form is ungrammatical, and will

never surface. For hoeden the Wrst candidate will be selected since it does

not violate any of the constraints, unlike the other candidates.

The potential asymmetry between morphological and phonological

structure can be expressed by alignment constraints. If the two types

of structure are to be isomorphic, the edges of stems have to be aligned

with the edges of phonological constituents such as the syllable. This is

what the constraints Alignment Left and Alignment Right require: align

the left and right morphological stem boundaries with phonological


constituent boundaries. The reason why Alignment Right is violated in the

case of hoeden has to do with an important universal phonological con-

straint mentioned above: syllables should, if possible, begin with a conson-

ant, the No Empty Onset constraint. This constraint refers to the notion

‘onset’ as a constituent of the syllable. Let me therefore introduce

here the basic notions of syllable structure. The following structure of

the syllable is usually assumed, illustrated in Figure 7.6 for the English

word stump.

Fig. 7.6 The syllable structure of stump

The asymmetry between phonological and morphological structure

observed with respect to hoeden shows that the No Empty Onset constraint

is ranked higher than Alignment Right. If Alignment Right ranked higher

than No Empty Onset, we would have to syllabify hoeden as hoed.en which

would result in the wrong phonetic form [hut.

e

n]. Ranking of No Empty

Onset above Alignment Right is therefore a partial speciWcation of the

interface between morphology and phonology in the grammar of Dutch.

We might be tempted to jump to a rash conclusion on the basis of the

facts discussed above: phonology cannot see the internal morphological

structure of words, and deals with complex words in the same way as it

deals with simplex words. Hence we do not need detailed speciWcations of

this interface. This conclusion is incorrect, however. A clear counterexam-

ple is that in many languages the morphological structure of compounds

plays an essential role in the computation of their phonetic forms, with

respect to both syllabiWcation and stress patterns. Consider the following

minimal pair of compounds of Dutch, with their syllabiWcation:

(4) [[bal]N[kanker]N]N ‘‘testicle cancer’’ (bal)�(kan)�(ker)�

[[balk]N[anker]N]N ‘‘beam brace’’ (balk)�(an)�(ker)�

The diVerence between these two compounds, which consist of the same

sequence of segments, is audible through their diVerent syllabiWcation

patterns. This is only possible if the syllabiWcation of compounds respects

160 interfaces

morphological structure. In particular, the requirement that the left bound-

ary of the second constituent aligns with a syllable boundary (expressed

by the constraint Alignment Left) is ranked higher than No Empty Onset.

Hence, in the second example, the second syllable begins with an empty

onset. Speakers of Dutch also notice this ranking through the eVect of

FinDevoicing: in a compound such as handappel ‘‘lit. hand-apple, eating

apple’’, with the morphological structure [[hand ]N[appel ]N]N and the

phonetic form [hAnt.A.p e

l], the /d/ is realized as [t], and hence it must be

located in coda position in order to be subject to devoicing. The example of

hoeden with the morphological structure [[hoed ]en], on the other hand,

shows that alignment of the right edge of a morphological stem

with a syllable boundary (Alignment Right) is less important than the No

Empty Onset Constraint. Hence we conclude to the following constraint

ranking (>> ¼ ‘‘ranked higher than’’):

(5) Alignment Left >> No Empty Onset >> Alignment Right

This ranking for Dutch (identical to the one for English) makes correct

predictions for preWxed words of Dutch, where the left stem boundary

usually coincides with a syllable boundary. The complex verb ver-as ‘‘to

incinerate’’ with the morphological structure [ver[as]N]V, for instance, syl-

labiWes in careful speech as ver.as, not as ver.as. This results in the second

syllable of this word having an empty onset.

In conclusion, morphological structure may aVect the computation of

phonological forms. Hence, the internal morphological structure of words

must be accessible to phonology, and the interface theory should specify in

which ways the morphological structure of a complex word determines its

phonetic form.

The representation of the phonological structure of the compounds in (4)

was in fact simpliWed for ease of exposition. The foot boundaries and

prosodic word boundaries were omitted. What should be added is that

each of the compound constituents consists of a phonological word of its

own (in their turn each consisting of one foot). Since the phonological word

is the domain of syllabiWcation, we will get the result that in a compound

the edges of its constituents coincide with phonological word boundaries.

Since the phonological word is the domain of syllabiWcation, this implies that

these morphological boundaries will also coincide with syllable boundaries.

The necessity to distinguish words in the morphological sense and phono-

logical words is a clear illustration of the asymmetry between phonology


and morphology: a morphological word may correspond with more than

one phonological word. In many languages this is the case for compounds.

SyllabiWcationpatterns areone typeof evidence for this, asdiscussedabove.

The domain of application of phonological rules or constraints is

another source of evidence for the prosodic structure of complex words.

In the case of Hungarian compounds, vowel harmony serves to determine

their prosodic structure. Vowel harmony is the phenomenon that all vowels

of a word share certain properties. In the case of Hungarian, the vowels of

a word are all either front vowels or back vowels. Front vowels are

articulated in the anterior part of the mouth, and back vowels in the back

part. Vowel harmony implies that many suYxes have two allomorphs, one

with a front, and one with a back vowel, as illustrated by the following

words (Siptar and Torkenczy 2000: 63):

(6) a. Buda-nak ‘‘Buda-dat’’, Pest-nek ‘‘Pest-dat’’

b. perd-ul- es-etek-tol ‘‘from your (pl) twirling around’’

ford-ul- as-otok-tol ‘‘from your (pl) turning around’’

In (6b) we observe four diVerent suYxes. Each of them has two allomorphs,

one with a front vowel, and one with a back vowel. The diVerence is

governed by the fact that the two roots of these words have a front vowel

/e/ and a back vowel /o/ respectively. (The acute accents on the vowel letters

of Hungarian orthography indicate length.)

The name Budapest for the capital of Hungary is a compound in which

the names of the two cities Buda and Pest have been combined. It seems

to violate the phonological constraint of vowel harmony because the Wrst

two vowels are back, and the last one is a front one. However, there is

no violation under the assumption that the domain of vowel harmony is

the phonological word, not the word in the morphological sense.

After having read (6a), you may wonder which is the correct dative suYx

for Budapest, -nek or -nak? After all, it is a suYx of the whole compound.

The correct one is -nek, since the dative suYx is a cohering suYx, and hence

forms one phonological word with the preceding material. We thus see here

another instantiation of the asymmetry between morphological and

phonological structure that we discussed above. Here are the two relevant

structures for Budapestnek:

(7) morphological structure: [[[buda]N[pest]N]Nnek]N

prosodic structure: (buda)ø (pestnek)ø

162 interfaces

As you can see, the vowels within each of the phonological words are

harmonic. In the Wrst phonological word all vowels are back, in the second

one they are all front.

The distinction between cohering and non-cohering suYxes is also relevant

for YidiÆ. In this Australian language, monosyllabic inXectional and

derivational suYxes are cohering, whereas disyllabic ones are non-cohering,

and form a phonological word of their own. We have to make that assump-

tion in order to account for the distribution of stressed and unstressed

syllables. Normally, words in YidiÆ begin either with a stressed or an

unstressed syllable, but always display an alternating pattern of stressed

and unstressed syllables. In complex words with disyllabic suYxes, however,

we Wnd both sequences of two unstressed and of two stressed syllables across

word-internalmorphological boundaries, since disyllabic suYxes begin a new

phonological word. The domain of stress assignment is the phonological

word, not the grammatical word. Hence, YidiÆ has words like the following

(Dixon 1977: 93).

(8) rhythmic pattern

waNal-muday ‘‘boomerang-comit.abs’’ (wa.Nal)ø (mu.day)ø

bigu:n-muda:y-Ædu ‘‘shield-comit-erg’’ (bi.gu:n)ø (mu.da:y.Ædu)ø

non-rhythmic pattern

waNal-muda:y-Ndu ‘‘boomerang-comit-erg’’ (wa.Nal)ø (mu.da:y.Ædu)øbigu:n-muday ‘‘shield-comit.abs’’ (bi.gu:n)ø (mu.day)ø

(The acute accents in these examples indicate stress.) The comitative suYx

in the second and third word has a lengthened vowel due to the presence of

stress. In the third word, both the second and the third syllable are

unstressed, whereas in the fourth word the second and the third syllable

are both stressed. The explanation for this disturbance of the rhythmic

alternation between stressed and unstressed syllables is that the domain of

this alternation is the phonological word. Since the suYxes used here are

disyllabic, they begin a phonological word of their own.

In Germanic languages we Wnd many non-cohering suYxes that derive

historically from lexemes, such as English -wise and -like. Although these

two suYxes have lost their lexical status, they still behave as phonological

words of their own, as can be concluded from the fact that they bear

secondary stress (money-wıse, book-lıke), just like the right constituents

of most English compounds. Some examples of non-cohering suYxes of

Dutch are:


(9) -achtig /Axt e

ª/ rood-achtig ‘‘reddish’’

-baar /ba:r/ eet-baar ‘‘edible’’

-dom /dOm/ adel-dom ‘‘nobility’’

-heid /hEid/ schoon-heid ‘‘beauty’’

Quite revealing is the contrast between the suYx -achtig and its semantic-

ally equivalent competitor, the cohering suYx -ig; both occur with the

adjectival stem rood ‘‘red’’, and contribute the same meaning ‘‘-ish’’, but

show diVerent phonological behaviour:

(10) rood-achtig [ro:t.Ax.t e

x] rod-ig [ro:.d

e

x]

Since -achtig forms a phonological word by itself, it is an independent

domain of syllabiWcation. Hence, the /d/ of rood occurs in syllable-Wnal

position, and is devoiced due to the constraint FinDevoicing. On the other

hand, the suYx -ig is cohering, and forms one prosodic word with its base.

Therefore, the morpheme-Wnal /d/ of rood Wlls an onset position in rodig,

and will be thus exempted from being devoiced.

Non-cohering aYxes are not necessarily phonological words. The Eng-

lish suYx -less /l

e

s/ is non-cohering, as we saw above. Yet, it does not form

a phonological word of its own, and cannot bear secondary stress since its

vowel is a schwa.

A characteristic of Dutch non-cohering suYxes that are phonological

words is that they allow for backward gapping. In that respect, they are like

compounds at the level of prosodic structure. Thus, they pattern with

compounds with respect to gapping: of two identical phonological words,

the Wrst can be deleted (Booij 1985):

(11) Gapping in compounds

land- en tuinbouw ‘‘agri(culture) and horticulture’’

wespen- en bijensteken ‘‘wasp (stings) and bee stings’’

hoofd- of nevenaccent ‘‘main (stress) or secondary stress’’

Gapping in suYxed words

storm- en regenachtig ‘‘storm(y) and rainy’’

zicht- en tastbaar ‘‘vis(ible) and tangible’’

christen- en heidendom ‘‘christian(ity) and heathendom’’

eenzijdig- of partijdigheid ‘‘onesided(ness) or partiality’’

A cohering suYx, on the other hand, cannot be gapped since it is not a

phonological word. For instance, the gapping of the cohering suYx -ig in

the phrase rodig en groenig ‘‘reddish and greenish’’ is impossible witness the

illformedness of *rood- en groenig.

164 interfaces

Similar things can be said about preWxes. PreWxes in Germanic languages

like Dutch with at least one full vowel often form a prosodic word of their

own, and hence such preWxed words are prosodic compounds, with the

concomitant stress pattern (main stress on the Wrst constituent, secondary

stress on the second):

(12) PreWx Example

aarts-/a:rts/ aarts-vıjand ‘‘arch-enemy’’

her-/hEr/ her-bebossing ‘‘reforestation’’

ex-/Eks/ ex-vrouw ‘‘ex-wife’’

anti-/Anti/ anti-betoging ‘‘anti-demonstration’’

PreWxes that are prosodic words of their own lend themselves easily for

being promoted to the status of lexeme. In English and Dutch, for instance,

ex can also be used nowadays as a word to denote one’s former partner.

This complies with the constraint that lexical words (non-function words)

must be well-formed prosodic words.

So far we have seen how the morphological and the prosodic structure of

words relate. This relationship also has an eVect on the phonological

makeup of aYxes. It is an old observation made by Roman Jakobson

that the phonological make-up of aYxes tends to diVer from that of lexical

morphemes. In Quechua, we Wnd a number of suYxes that begin with

consonant clusters that are never found at the beginning of lexical mor-

phemes, as is illustrated by the following word (van de Kerke 1996: 126):

(13) maylla-wa-rqa-nki-ku

wash—1obj-past-2sg-pl

‘‘You have washed us’’

Since both the lexical root and the suYxes end in a vowel, clusters such as /rq/

and /nk/ at the beginning of a suYx will be split up in the prosodic structure,

and assigned to diVerent syllables. The syllabiWcation of the example (13)

will be mayl.la.war.qan.ki.ku, and hence creates no phonotactic problems.

Words must indeed be pronounceable on their own, unlike aYxes. That is,

words need to comply with the requirements on well-formed phonological

words. Dutch requires phonological words to have at least one full vowel

(that is, a vowel that is not the schwa). The schwa is a special vowel in that it

cannot bear stress, and hence creates an unstressable syllable. Consequently,

it is never the case that the only vowel of a Dutch word is the schwa (with the

exception of function words such as determiners). On the other hand, there


are many Dutch suYxes with schwa as their only vowel, such as -er /

e

r/.

This is to be expected, as suYxes will not appear as phonological words

themselves (note that the non-cohering suYxes listed in (9) all contain a full

vowel). A related remarkable property of suYxes is that they may consist of

consonants only, again unlike words. In Germanic languages most conson-

antal suYxes consist of /s/, /t/, or a combination thereof, consonants that

can freely occur at the end of word-Wnal syllables. To conclude, we can make

the generalization that the diVerent subclasses of morphemes of a language

(lexical morphemes, preWxes, suYxes, etc.) are phonologically shaped in such

a way that they can lead to phonologically well-formed words.

This relationship between morphology and phonology will also help us

to understand the notion clitic. Clitics are ‘small words’ of functional, non-

lexical categories such as pronouns and determiners that ‘lean on’ (the

word clitic derives from the Greek verb klinein ‘‘to lean’’) a preceding or

following host word, and cannot appear as phonological words by them-

selves. For instance, in Italian lexical morphemes have to consist of at least

two syllables (there are a handful of exceptions such as re ‘‘king’’). Most

pronouns, however, are monosyllabic. These clitics therefore take an adja-

cent word as their host, and form one prosodic constituent with that word.

The Italian clitic pronouns precede their host word (this is called proclisis),

except in the case of inWnitival and imperative verbal forms that require the

clitics to follow them (enclisis):

(14) proclisis: me lo racconta

me it tell.3sg

‘‘(s)he tells me it’’

enclisis: racconta-me-lo

tell.imp-me-it

‘‘tell me it’’

Clitics therefore share a property with cohering aYxes: their phonological

dependence on a host. In this respect, proclitics are like preWxes, and

enclitics like suYxes. Words in the morphosyntactic sense do not always

correspond in a one-to-one fashion to words in the phonological sense.

Clitics form another illustration of this asymmetry between morphosyntac-

tic and prosodic structure, and the study of clitics belongs (partially) to the

study of interface between phonological and morphosyntactic structure.

The similarity in phonological behaviour of aYxes and clitics makes it

sometimes diYcult to determine if a morpheme is to be considered an aYx

or a clitic. Consider the following examples from English:

166 interfaces

(15) a. the king of England’s hat

b. the boy across the road’s cycle

c. the man I talked about’s car

The morpheme s used here is historically a genitive suYx, but it has

developed into a clitic that can be attached at the end of the possessor

phrase. Hence, it is sometimes called a phrasal aYx. Whereas suYxes are

attached to words of particular categories, the morpheme s attaches not

only to nouns but to whatever word happens to occur in phrase-Wnal

position, so even to a preposition such as about, as in (15c).

Other examples of suYx-like clitics might be claimed to exist in Icelandic.

In this language, the deWnite article may be ‘suYxed’ to the preceding noun.

It can also occur as a free article before a noun preceded by an adjective.

Both the noun and the ‘suYxed’ deWnite article are inXected for number

and case. Consider the following singular forms of the word hestur ‘‘horse’’

given in Table 7.1. As you can see, both the noun and the deWnite article are

inXected for case. The bound forms of the article diVer from the full forms

in that the initial /h/ is omitted. A possible interpretation of these facts is

that the article is an enclitic when occurring after the noun.

The diVerence between aYxes and clitics can sometimes be seen in the

diVerential eVect that they have on the phonetic form of words. Dutch has

a clitic pronoun er /

e

r/ ‘her’ that is attached prosodically to a host word on

its left. This causes resyllabiWcation of the word þ clitic sequence, just like

vowel-initial cohering suYxes do. The diVerence is that attachment of a

suYx pre-empts application of Wnal devoicing to stem-Wnal obstruents,

whereas clitics with a similar phonological form do not, as shown by the

following minimal pair in (16):

(16) vind-er ‘‘Wnd-er’’ [vind-

e

r] vs (ik) vind er [vin.t

e

r] ‘‘(I) Wnd her’’

That is, Wnal devoicing must have applied to the Wnite verb form vind before

the clitic is attached. Therefore, clitics are sometimes called postlexical aYxes

Table 7.1. The singular paradigms of hestur and the deWnite article

free article noun noun þ suYxed article

nom hin-n hest-ur hestur-inn

acc hin-n hest hest-inn

dat hin-um hest-i hesti-num

gen hin-s hest-s hests-ins

Source: Thrainsson 1994: 156.


since they have to be attached after the rules of word phonology (the lexical

phonology) of the language have applied. The attachment of the clitic er

triggers resyllabiWcation, because the clitic forms one prosodic word with

the preceding verb. Hence, the /d/ of vind Wrst devoices in coda position,

and subsequently, at the postlexical level, it moves to an onset position.

7.3 Allomorphy and affix competition

Morphemes may exhibit variation in their phonological shape. This vari-

ation in shape may have nothing to do with phonology. This is the case

when languages have morphological systems with more than one stem-

form; each stem-form has to be used for particular inXectional categories

(Chapter 6). Allomorphy may also be a completely predictable eVect of

phonology, as shown in the previous section for the allomorphy related to

Wnal devoicing in Dutch. In other cases, the phonological alternations are

regular too, but apply to a restricted set of words only. For instance, the

Dutch diminutive suYx has Wve allomorphs, whose distribution is predict-

able (the letter e stands for a schwa, and ng indicates the velar nasal):

(17) a. -je after stem-Wnal obstruents;

b. -etje after sonorant consonants preceded by a short vowel with primary or

secondary stress;

c. -pje after stem-Wnal /m/ except in cases falling under b;

d. -kje after stems ending in the velar nasal /N /;

e. -tje elsewhere

This allomorphy is illustrated by the words in (18):

(18) base noun diminutive

a. lip ‘‘lip’’ lip-je

hek ‘‘gate’’ hek-je

b. ring ‘‘ring’’ ring-etje

serıng ‘‘lilac’’ sering-etje

c. riem ‘‘belt’’ riem-pje

bez[

e

]m ‘‘sweep’’ bezem-pje

d. koning ‘‘king’’ konin-kje

paling ‘‘eel’’ palin-kje

e. ree ‘‘deer’’ ree-tje

traan ‘‘tear’’ traan-tje

168 interfaces

The diVerent allomorphs of the diminutive suYx might be derived from an

underlying form /tj

e

/ that surfaces in (18e), by assuming rules that delete the

/t/ (18a), insert a schwa (18b), or assimilate the /t/ to the place of articulation

of the preceding nasal (18c–d). However, these rules apply exclusively to

diminutive nouns. For instance, there is no general rule of schwa insertion in

the context mentioned in (17b). The Dutch complex word stil-te ‘‘silence’’,

for instance, is not realized as [stØl

e

t

e

]. Hence, these rules must mention

the morphological property [þdiminutive] in their structural description.

This means that they are morphologically conditioned phonological rules.

In other cases, the rule applies to a Wxed set of lexical items. That is,

the alternation is lexically conditioned. This applies to a number of mono-

syllabic Dutch nouns: in the plural form (and some derived words as well)

the stem vowel is lengthened, which is a relict of the process of Open

Syllable Lengthening (OSL) that was active in Early Germanic:


d[A]g ‘‘day’’ d[a:]gen

h[O]f ‘‘court’’ h[o:]ven

w[E]g ‘‘road’’ w[e:]gen

If one assumes a phonological rule for such alternations, the relevant nouns

have to be marked as undergoing this rule in the lexicon as [þOSL], and

this diacritic feature has to be mentioned in the structural description of the

rule as well. So this rule may be formulated as follows:

(20) V ! V: in the context —)�

[þOSL]

The word dag, for instance, will be marked as [þOSL] in the lexicon, and

hence all its segments carry that diacritic feature.

It might also be the case that a rule is governed both by morphological

and by lexical features, as is the case for German umlaut. This is the process

in which the back vowels of stems are fronted before certain suYxes that

originally contained front vowels or glides. As the examples (21a) show,

application of umlaut depends on the individual stem (Wiese 1996: 188):

(21) a. Umlaut No umlaut

Vater ‘‘father’’–Vater-chen ‘‘dim’’ Onkel ‘‘uncle’’–Onkel-chen ‘‘dim’’

laufen ‘‘to walk’’–lauf-t ‘‘3sg’’ rufen ‘‘to call’’–ruf-t ‘‘3sg’’

b. Hund ‘‘dog’’–Hund-in ‘‘dog, fem’’ Hund-e ‘‘dog, pl’’, Hund-chen

Hund-chen ‘‘dog, dim’’ ‘‘dog, dim’’


The examples Hunde and Hundchen in (21b) show that lexical items may

have to be marked as exceptions to umlaut. In the case of the diminutive

noun for Hund, the umlauted form is the standard one, but the other one

also occurs. The lexical stem Hund does allow for umlaut, as in Hun-din,

and the plural suYx and diminutive suYx both trigger umlaut. Yet there is

no umlaut in the plural form, and umlaut is optional in the diminutive

form. A useful cover term for such morphologically and/or lexically

conditioned phonological rules is morpholexical rule.

This is not the whole story about allomorphy, however. Consider the

following pairs of related words in English:

(22) drama, dramat-ic, dramat-ist

Plato, platon-ist, platon-ism

In these examples, the basewords dramaandPlatohave a short form,whereas

stem-forms with an additional consonant are used for derivation. This is a

reXex of the history of these originally Greek words: the long form is the

underlying form, but the stem-Wnal consonant was dropped in nom.sg forms.

The eVect for present-dayEnglish is that the long form is tobeused for suYxes

of the non-native learned stratum, whereas the short form is to be used before

native, Germanic suYxes, and with preWxes. For instance, the plural form of

drama is dramas, not dramats. Thus, the distribution of these allomorphs is

stated in morphological terms. It is hard to see how we might provide an

insightful account of this kind of allomorphy in terms of phonological rules.

Allomorphy as a reXex of history is also found in word pairs such as the

following:

(23) deduce deduct-ion

induce induct-ion

produce product-ion

reduce reduct-ion

This allomorphy is a reXex of the system of diVerent stems for Latin verbs:

derivation takes the participial stem form of the Latin verb ducere, duct, as

its base (cf. section 6.3).

Historically determined allomorphy is also found in the formation of

French de-adjectival adverbs in -ment, see (24). The feminine forms of the

adjectives in (24) are irregular, except the Wrst one. The data show that

the suYx -ment takes the feminine form of the adjective as stem for adverb

formation. Note, however, that there is no feminine meaning involved

170 interfaces

in the meaning of the adverb. This is a case of paradigmatically governed

allomorphy: the correct form of the adjectival stem is determined by

referring to that of a paradigmatically related form, the feminine form.

Historically, this is the eVect of the suYx -ment. It derives from the

ablative form of the Latin feminine noun mens ‘‘mind’’ that required the

feminine form of the preceding adjective, as in clara mente ‘‘clear-abl.sg

mind-abl.sg., with a clear mind’’.

(24) masc adj. fem adj. Adverb

lent ‘‘slow’’ lente lente-ment

beau ‘‘beautiful’’ belle belle-ment

vert ‘‘green’’ verte verte-ment

fou ‘‘mad’’ folle folle-ment

The examples discussed so far are cases of stem allomorphy. This kind of

allomorphy is also found in compounds where the Wrst constituent may have

a form that cannot be used when it is used as a word of its own. Wellknown

cases are the linking elements in Germanic languages, illustrated here for

Swedish (Anderson 1994: 277):

(25) land-s-ting ‘‘city council’’, las-e-bok ‘‘textbook’’, kvinn-o-arbete ‘‘women’s

work’’

Some of these linking elements are old genitive endings. Noun phrases with

an inXected modifying noun became compounds, and thus the inXectional

ending got trapped in between the two constituents:

(26) [N-gen N]NP > [N-linking element-N]N

Subsequently, such compounds served as models in the coining of new

compounds that, by analogy to existing compounds, got these linking

elements as well. Phonologically, these linking elements form one phono-

logical word with the Wrst constituent of compounds, as we can conclude

fromword-internal gapping inDutch compounds such as the following one:

(27) wespen-en bijen-steken ‘‘wasp stings and bee stings’’

[[wesp]Nen[steken]N]N en [[bij]Nen[steken]N]N

(wes.pen)ø(ste.ken)ø(en)ø(bij.en)ø(ste.ken)ø

This example shows that the linking element is not included in the gapping

and belongs to the Wrst phonological word of compounds, as is also shown

by the prosodic structure of the non-gapped compounds in (27).


Let us now have a look at another kind of allomorphy, aYx allomorphy.

Dutch has two counterparts to the English deverbal and denominal suYx

-er: the suYxes -er and -aar. These two suYxes look like allomorphs in

the sense that they are phonologically similar. However, it is not possible

to assign them a common underlying form, and derive the two surface

forms by means of well-motivated general phonological rules or constraints

of Dutch. There is no general phonological constraint for Dutch that

vowels in word-Wnal unstressed syllables must be reduced to schwa.

Hence, -er cannot be derived phonologically from -aar. So in fact you

can see them as competing aYxes: diVerent aYxes with the same meaning

and domain of application. When the degree of phonological similarity

between competing aYxes is considerable, they may be considered

allomorphs of one morpheme. When the similarity is less, it is better

to regard them as distinct morphemes, and thus in a relationship that

resembles the suppletive one between the stem-forms good and bet- of the

lexeme good.

The basis for choosing between the suYxes -er and -aar is the following:

-aar is used after a stem ending in an unstressed syllable, with /l/, /r/, or /n/

as its Wnal consonant (these are the coronal sonorant consonants of Dutch);

-er is used elsewhere.

(28) bedel/be:d

e

l/ ‘‘to beg’’ bedel-aar /be:d

e

la:r/ ‘‘begger’’

luister /lVyst

e

r/ ‘‘to listen’’ luister-aar /lVyst

e

ra:r/ ‘‘listener’’

reken /re:k

e

n/ ‘‘to compute’’ reken-aar /re:k

e

na:r/ ‘‘computer’’

bezem /be:z

e

m/ ‘‘to sweep’’ bezem-er /be:z

e

m

e

r/ ‘‘sweeper’’

verdedig /vErde:d

eg/ ‘‘to defend’’ verdedig-er /vErde:d

eeg e

r/ ‘‘defender’’

bak /bAk/ ‘‘to bake’’ bakk-er /bAk e

r/ ‘‘baker’’

It is possible to account for this pattern by assigning a phonological

subcategorization feature to the suYx -aar that speciWes its restricted

distribution (only after stem-Wnal coronal sonorants preceded by a schwa).

The suYx -er does not need to have such a subcategorization feature: it is

the default suYx that can be used elsewhere. The priority of -aar above -er

for verbs ending in an unstressed syllable with /l, r, n/ can be taken care of

by Panini’s Principle introduced in Chapter 5. Recall that this principle

states that if two or more rules compete, the more speciWc rules get priority

over the less speciWc rules, and that the application of the more speciWc rule

pre-empts application of the more general one. In the case of the base word

bedel the suYx -aar will be selected, and hence suYxation with -er is

172 interfaces

excluded. The base word bezem, on the other hand, although it ends in an

unstressed syllable, does not end in a coronal sonorant since the /m/ is a

labial consonant. Thus, attachment of -aar is impossible, and -er will be

attached.

This analysis can be qualiWed as an analysis that makes use of input

constraints: each competing suYx imposes certain requirements on its

input forms. The drawback of such an analysis is that it does not provide

any explanation for this particular selection pattern. The factor behind this

pattern is most probably the avoidance of a sequence of two unstressed

syllables. If we added -er after an unstressed syllable, we would get

a sequence of two unstressed syllables. Germanic languages strive for a

parsing of their words into disyllabic trochees. That is, ideally, the prosodic

form of each word consists of one or more trochees. In addition, there can

be monosyllabic feet if the relevant syllable has a full vowel, and is long

enough. Consider the word reken-aar and its ill-formed counterpart reken-

er. Their prosodic structure in terms of syllables and feet will be as follows

(syllable boundaries indicated by dots):

(29) a. (re:.k

e

)F(na:r)F

b. (re:.ke

.)Fne

r

In (29b), the last syllable cannot be made part of a foot (that is, cannot be

parsed into a foot), because it does not have a full vowel. Hence, (29a) is

more optimal from the point of view of prosodic structure. This prosodic

structure correctly implies that the last syllable also bears a degree of stress,

i.e. secondary stress.

This selection process of competing allomorphs can be modelled in

terms of the optimality-theoretical tableaux introduced in section 7.2:

each combination of a stem and an allomorph is a candidate, and the ranked

set of phonological constraints will determine the optimal candidate.

The advantage of this theory is that it makes use of output constraints.

The crucial constraints involved here are ParseSyll, Foot-Min, and Foot-

Max. ParseSyll requires syllables to bemade part of feet, Foot-Max requires

feet to be maximally binary, and Foot-Min requires feet to be minimally

binary. The computation is shown in Figure 7.7.

As you can see, the ranking of ParseSyll and Foot-Max is not crucial in

these cases. What is crucial is that these two are ranked higher than Foot-

Min. As a result of this ranking, the Wrst candidate is designated as the

optimal one.


In the analysis outlined here, the selection of these two competing suYxes

is done by output constraints. Such constraints thus play an important role

in the interface between morphology and phonology: morphology

provides a number of alternatives, equivalent from the morphological

point of view, and the phonology then computes which of them is optimal

from the phonological point of view.

This does not mean that phonological subcategorization constraints are

superXuous. For instance, we still have to state that -aar can only occur

after stems ending in a coronal sonorant consonant (that is, /l/, /r/, or /n/).

Therefore, the deverbal noun for bezem ‘‘to sweep’’ is bezemer [be:.z

e

.m

e

r]

even though this deverbal noun ends in a sequence of two unstressed

syllables. Recall also that these constraints are all violable. For instance,

Dutch diminutives such as ball-etje [bAl etj e] ‘‘ball, dim’’ and kamm-etje

[kAm e

tj

e

] ‘‘comb, dim’’ end in a sequence of two unstressed syllables, and

hence ParseSyll will be violated. In this case there is no morphological

alternative that can be used to avoid this violation. In other words, when

necessary, morphology takes precedence over phonology, even if this leads

to less optimal phonological structures.

Another example of the role of prosody in allomorph selection comes

from Biak, a language spoken in NewGuinea. This language uses either the

inWxes -w- and -y- or the preWxes wa- and i- for 2sg and 3sg forms of the

verb respectively (Table 7.2). The choice between inWx and preWx is lexically

governed, as illustrated by (a–b). In the b-examples, we might have

expected the inWxed forms rwov and ryov, but these are not the forms

used. However, if the verbal stem begins with a consonant cluster, it is

always the preWx allomorph that has to be chosen since an inWx would lead

to phonotactically impossible consonant sequences (c–e).

In conclusion, even if allomorphs cannot be derived from a common

underlying form, it is nevertheless possible to make phonological general-

izations concerning their distribution.

Fig. 7.7 OT-tableau for rekenaar

174 interfaces

7.4 Cyclicity and co-phonologies

In many languages, the stress patterns of complex words are, at least

partially, determined by their morphological structure. All Germanic lan-

guages exhibit this kind of interface between morphology and phonology.

A simple example is the way in which the stress patterns of compounds

have to be computed. The English compound sugar cookie has primary

stress on its Wrst syllable, and secondary stress on its penultimate one: sugar

cookie. This stress pattern can be computed as follows. First, we assign

stress to each constituent. Both words have penultimate stress. The second

step is to determine which of these two stresses is prominent. The rule

for English nominal compounds is that the stressed syllable of the Wrst

constituent has the highest prominence. A traditional way of formalizing

this type of computation is to assume that phonological rules of stress

assignment apply cyclically, in an outward fashion where one begins with

the smallest constituents:

(30) [[sugar] [cookie]]

1st cycle, Main Stress Rule __1___ __1____

2nd cycle, Compound Stress Rule __1_______2____

On the Wrst cycle, theMain Stress Rule of English assigns primary stress to

one of the vowels of each constituent. On the second cycle, the Compound

Stress Rule reassigns primary stress to the Wrst primary stressed vowel, and

causes all other stresses to be automatically lowered by one degree.

This formal machinery, from Chomsky and Halle (1968), is outdated by

now as far as the analysis of stress patterns is concerned, but serves well to

grasp the idea of cyclic application of rules. In contrast to Chomsky and

Halle’s analysis, stress is considered nowadays as a property of prosodic

Table 7.2. Allomorph selection in Biak

stem 2sg 3 sg gloss

a. ra rwa rya to go

b. rov warov irov to Xy

c. kvok wakvok ikvok to stand up

d. snai wasnai isnai to be clear

e. smai wasmai ismai to have

Source: W. van den Heuvel, pers. comm.


constituents. Hence, the inXuence of morphological structure on stress

patterns of complex words is indirect. First, we map morphological

structure onto prosodic structure, and the prosodic constituents are

marked for stress. One way of doing this is to use the labels ‘strong’ (s)

and ‘weak’ (w). The prosodic structure of sugar cookie will hence look as

in Figure 7.8.

Fig. 7.8 The prosodic structure of sugar cookie

Strong–weak is a relation between two sister constituents. One of them is

the head of a prosodic category such as the foot. The Wrst syllable of sugar

cookie is the strongest syllable of the strongest phonological word, and

hence it carries the primary stress of this compound. The strongest syllable

of the weak phonological word, coo, has secondary stress. In short, we do

not need cyclicity as a special principle of rule application. The interface

between morphology and phonology discussed here is fully covered by the

following two mapping principles for English:

(i) each constituent of a compound corresponds with a phonological

word;

(ii) in the case of N þ N compounds, the Wrst phonological word is the

strongest.

This approach also enables us to give an adequate account of the stress

patterns of complex words with non-cohering suYxes. This can be illus-

trated by means of the Dutch word draagbaar that has two meanings:

(31) compound [[draag]V[baar] N]N ‘‘stretcher’’

derived word [[draag]Vbaar]A ‘‘port-able’’

The suYx -baar is one of the non-cohering suYxes of Dutch. Therefore,

from the prosodic point of view both words are prosodic compounds, and

have the same prosodic structure (draag)ø (baar)ø. The Wrst syllable carries

main stress, the second one secondary stress, as is represented correctly by

this prosodic structure. The same prosodic status holds for many preWxes of

176 interfaces

Greek and Latin origin in Germanic languages that also carry stress, as

illustrated here for English:

(32) a. main stress on preWx: counter-argument, sub-set

b. secondary stress on preWx: anti-relıgious, neo-classicısm,

Another aspect of the role of morphological structure in the computation of

the phonetic forms of complex words is the phenomenon of co-phonologies.

As we saw in section 3.3, the morphological system of a language may

consist of more than one stratum. Each stratum may have its own phono-

logical system, that is, its own co-phonology. In Germanic languages, for

instance, words that end in non-native suYxes are stressed in the same way

as simplex words. Hence, they might carry the main stress of the word.

SuYxes of Germanic origin, on the other hand, are mostly stress-neutral:

the addition of such a suYx does not shift the location of main stress

rightward. This is illustrated here for Swedish (Andersson 1994: 278):

(33) Non-native suYxation, stress on last syllable: brygg-eri ‘‘brewery’’,

individual-itet ‘‘individuality’’, dans-or ‘‘dancer’’, or on penultimate syllable:

prost-inna ‘‘dean’s wife’’;

Native suYxation, stress on Wrst syllable: sok-ande ‘‘search’’, bo-ende ‘‘living’’,

las-ning ‘‘reading’’, bak-else ‘‘pastry’’

These examples once more show that phonology makes use of morpho-

logical information concerning words.

7.5 The morphological use of phonology

The interface phenomena discussed so far have all been cases in which

phonology makes use of morphological information. The inverse situation,

morphology making use of phonology, is also found in natural languages.

Simple examples are cases in which the use of a particular aYx is

conditioned phonologically. The English comparative suYx -er, for

example, can only—with some minor exceptions—be attached to mono-

syllabic adjectival stems, and to disyllabic stems ending in a light syllable.

In other cases, a periphrastic form with more has to be used.

(34) green greener

silly sillier

obese *obeser / more obese

excellent *excellenter / more excellent


Dutch is a language in which prosodic considerations play a role in the

extent to which aYxes can be stacked up in complex words. In the case of

preWxation, a sequence of two unstressed preWxes will be avoided. Recall

that Dutch parses words in trochees (section 7.1). Hence, a sequence of two

unstressed preWxes at the beginning of a morphological word will violate

the condition ParseSyll twice. Stressed preWxes, on the other hand can be

preWxed to preWxed stems. The contrast is shown in (35):

(35) ont-bos ‘‘to deforest’’ *ver-ont-bos ‘‘to destroy through deforestation’’

on-geluk ‘‘accident’’ ver-on-geluk ‘‘to die in an accident’’

The past participles of Dutch verbs are formed through circumWxation:

the preWx ge- and the suYx -t/d (regular verbs) or -en (irregular stem-

alternating verbs). However, if the verbal stem begins with an unstressed

preWx, the participial preWx is obligatorily omitted. Thus, a sequence of

unparsable syllables is avoided once more:

(36) Verbal stem Past participle

vang ‘‘catch’’ ge-vang-en

ont-vang ‘‘receive’’ ont-vang-en

stuur ‘‘send’’ ge-stuur-d

be-stuur ‘‘govern’’ be-stuur-d

pre-Wgeer ‘‘preWx’’ ge-pre-Wgeerd

re-animeer ‘‘reanimate’’ ge-re-animeer-d

The native preWxes ont- and be- do not receive stress at all; the non-native

preWxes pre- and re- receive rhythmic secondary stress. Hence, the latter

preWxes do not block the attachment of the participial preWx ge-.

The morphological use of phonological patterns and categories is a prom-

inent feature in reduplication. Consider the cases in (37) of partial reduplica-

tion in the perfect tense forms of Latin and Greek (Wiltshire and Marantz

2000: 563). In these examples, the perfect stem is formed by preWxing a copy

of a part of the base; the copy is called the reduplicant. Recall that this kind

of reduplication is called partial reduplication because it is not the whole

word that is copied. An example of full reduplication is given in section 2.2.

A Wrst, informal deWnition of the Latin reduplication process is ‘copy the

Wrst syllable of the base’. An alternative formalization is to say that Latin

has a phonologically underspeciWed reduplicative preWx with the abstract

form CV, a sequence of a consonant and a vowel. The sound segments of

the base are then linked to the CV templates, from left to right. Since there

are only two positions, only the Wrst two segments of the base will be

178 interfaces

(37) Base Perfect stem Gloss

Latin curr- cu-curr- run

dic- di-dic- learn

mord- mo-mord- bite

pend- pe-pend- hang

Greek lu- le-lu- wash

ly- le-ly- loose

graph- ge-graph- write

klin- ke-klin- incline

copied, as illustrated in (38) for the perfect stem pepend-. This is the analysis

proposed in Marantz (1982). It can be qualiWed as autosegmental morph-

ology, since in this kind of analysis the abstract CV-tier of phonological

structure, which is an autonomous layer of phonological structure, has a

morphological role: the CV sequence has a morphological function of its

own. The morphological use of abstract phonological tiers is also exem-

pliWed in section 2.2, in the analysis of the non-concatenative or root-and-

pattern morphology of Hebrew verbal forms.

(38) C V C V C C

j j j j j jp e n d p e n d

The Greek examples diVer from the Latin ones in that only the initial

consonant of the base is copied. That is, the vowel in the reduplicative CV

preWx is a Wxed one, e. Thus, the V of the Greek CV-preWx is already linked

to a vowel /e/ on the segmental tier.

The following examples from Ilokano, a language of the Philippines,

show that the reduplicative aYx of this language cannot be speciWed as

‘‘copy the Wrst syllable’’, but has to be speciWed in terms of a constant,

abstract phonological shape (McCarthy and Prince 1998: 285):

(39) kaldı ‘‘goat’’ kal-kaldı ‘‘goats’’

pusa ‘‘cat’’ pus-pusa ‘‘cats’’

In the Wrst example, the reduplicant is identical to the Wrst syllable. In the

second example, however, this is not the case, since the Wrst consonant of

the second syllable is also copied. So we have to conclude that the redupli-

cative aYx has to be speciWed as a heavy (or bimoraic) syllable, that is, a


syllable that either ends in a consonant, or has a long vowel. This kind of

syllable is referred to as a bimoraic syllable: both a long vowel (VV) and a

VC sequence count as two morae.

The Ilokano example serves to illustrate the basic idea of prosodic

morphology which claims that the shape of reduplicative aYxes has to be

deWned in terms of prosodic categories such as the syllable or the prosodic

word. In the Latin case, the reduplicative preWx can be deWned as a light

syllable (that is, a CV syllable), and in Ilokano as a heavy syllable. Another

possible form of the reduplicant is that of a minimal prosodic word.

The minimal prosodic word of a language is often a disyllabic foot.

Consider the following examples from the Australian language Diyari,

where reduplication serves a number of morphological functions (Austin

1981; cited in McCarthy and Prince 1998: 286–7).

(40) kanku-kanku ‘‘boy’’

kulku-kulkuNa ‘‘to jump’’

tjilpa-tjilparku ‘‘bird species’’

In these examples, the reduplicant consists of two syllables, which form a

minimal prosodic word. The segments of the base are mapped onto this

abstract prosodic pattern, and thus maximally two syllables of the base are

copied.The third example is quite interestingbecause the last consonantof the

second syllable is not copied. The reason for this is that an open syllable is

the unmarked form of a syllable, and thus preferred. In the Wrst syllable of the

reduplicant, on the other hand, we Wnd a closed syllable, a more marked

syllable structure. We need this closed syllable because otherwise we would

have to skipa segment of the base in themappingof segmental structure to the

prosodically deWned reduplicative preWx. So it is only the second syllable that

canhave theunmarkedformofanopensyllable.Theappearanceofunmarked

structure in this kind of morphological operation is called ‘the emergence of

the unmarked’. It can be nicely modelled bymeans of Optimality Theoretical

tableaux. We need the following, informally circumscribed constraints:

(41) Reduplicant ¼ Minimal Prosodic Word

Open Syllable: a syllable is open

No Skipping

If we assume that No Skipping is ranked higher than Open Syllable, we see

that of the following candidates, the Wrst is the optimal one. This is shown

in Figure 7.9. The second and the fourth candidate violate one of the two

180 interfaces

highest ranked constraints, and are therefore ill-formed. The third candi-

date violates the constraint Open Syllable four times, whereas the Wrst

candidate does this only three times. Hence, the fourth violation of Open

Syllable by the third candidate is fatal. Therefore, the Wrst candidate is

qualiWed as the optimal one.

The morphological use of the prosodic category ‘minimal prosodic word’

is also found in the formation of hypocoristics (endearment forms of

proper names) through truncation (see section 1.4), as the following

examples illustrate (Lappe 2002: 135):

(42) German Andreas Andi

Dagmar Daggi

Spanish Alexandrina Dina

Ernesto Neto

French Dominique Dom

Valerie Val

English Patrick Pat/Patty

Elizabeth Liz/Lizzy

In German and Spanish the form of these names is that of a disyllabic foot,

in French the form is that of a heavy syllable, and English has both options.

Both disyllabic feet and heavy syllables can function as minimal prosodic

words.

A last example of the use of phonology in morphology is the use of tonal

morphemes, that is, morphological operations that use tones to signal a

particular morphological category. The Limburgian dialect of Maasbracht

(The Netherlands) is an example of a language that makes use of tone

contrasts for morphological purposes. The contrast is that between a

dragging tone (High–Low–High) versus a falling tone (High–Low), either

of which occurs on the stressed syllable of a word. For instance, the neuter

and feminine forms of a number of adjectives diVer in that the neuter form

tjil.pa.tjil.par.ku

/RED + tjilparku/ RED=MINWORD NO SKIPPING OPEN SYLLABLE

tji.pa.tjil.par.ku

tjil.park.tjil.par.ku

tji.tjil.par.ku *!

**** !

***F

*! **

**

Fig. 7.9 Reduplication in Diyari


has a dragging tone, and the feminine form a falling tone, see (43). This is

another example of autosegmental morphology since it is the tonal tier that

is involved in expressing the morphological categories involved, whereas

the segmental tier remains unchanged.

(43) Neuter, HLH Feminine, HL Gloss

wiis wiis wise

stiif stiif stiV

ªriis ªriis grey

Summary

This chapter focused on the interface between morphology and phonology.

Morphological structure appeared to inXuence the phonetic forms of com-

plex forms through principles of alignment that require phonological

boundaries to coincide with morphological ones. Yet, this alignment is

not perfect, and there are many cases of asymmetry between morphological

and phonological structure. A second form of interaction is the choice

between stem allomorphs or competing aYxes. This choice may be gov-

erned by considerations of phonological optimality. In some cases, the

choice can be insightfully modelled by means of output conditions, as in

Optimality Theory.

The non-isomorphy between prosodic structure and morphological

structure is also the key to the understanding of the phonological similar-

ities between aYxes and clitics.

Morphologymakes use of phonology in a number ofways.Morphological

operations may impose phonological constraints on the stems they

operate upon. Phonological processes such as copying and truncation

have morphological functions. Prosodic morphology makes use of

prosodic categories such as the syllable and the prosodic word for

morphological operations.

Questions

1. Show that the following English suffixes must be considered as cohering:-able, -er, -ing.

182 interfaces

2. The Dutch determiners de ‘‘the’’ and een ‘‘a’’ have the phonological forms/d e/ and / en/ respectively. How can we tell from these phonological formsthat these words must be function words?

3. Why would one prefer to consider the allomorphy that can be observed innoun–adjective pairs such as Plato–Platonic and realist–realıstic (-ic/-nic) asa case of stem allomorphy, rather than as a case of suffix allomorphy?

4. The Italian negative prefixes in- and non- differ in that it is only in- thatundergoes assimilation to a following stem-initial consonant:

ir-regolare ‘‘irregular’’, im-maturo ‘‘immature’’non-lavabile ‘‘non-washable’’, non-memorizzable ‘‘non-memorizable’’

How might this difference in phonological behaviour be explained?

5. Show that the syllabification pattern of English words with the suffix -ercomplies with the constraint ranking in (5).

6. Consider the following case of gapping in Dutch (the parenthesized partberen of the compound ijsberen is gapped):

ijs(beren) en bruine beren ‘‘lit. ice and brown bears, polar bears and brownbears’’

How can this case of gapping be used to show that gapping in Dutch isconditioned prosodically, and not syntactically?

7. The Dutch derivational suffix -tig/t ex/, used in number names such astwin-tig ‘‘twenty’’ and der-tig ‘‘thirty’’, has developed into a numeralwith the meaning ‘‘umpteen’’. In this use, the word is pronounced as[tØx]. How can this phonological change be explained?

8. Suppose that someone claimed that the stress pattern of the Englishcompound breakfast table can be computed without reference to mor-phological structure, by means of the following rule: in native words mainstress falls on the first syllable, and secondary stress on every odd syllablegoing from left to right. Provide empirical evidence that speaks againstsuch an analysis, and in favour of the role of morphological structure incomputing the stress patterns of compounds.

9. Consider the following Hungarian words and their dative forms:

Vajda-ne ‘‘wife of Vajda’’ Vajda-ne-nak (dative form)Szoke-ne ‘‘wife of Szoke’’ Szoke-ne-nek (dative form)

Which conclusion can you draw about the role of the suffix -ne in vowelharmony?

10. Which prosodic constraints can be invoked to explain the choice betweenthe English definite articles a and an?


Further reading

A classic article on the interface between phonology and morphologyin Optimality Theory is McCarthy and Prince (1994). Cyclic application ofphonological rules is one of the building stones of the theory of LexicalPhonology; see Booij (2000) for an overview of this theory.Criteria for distinguishing clitics from affixes are given in Zwicky and Pullum

(1983). Clitics often have a special syntactic distribution as well, cf. vanRiemsdijk (1998) for studies on clitics in European languages. The necessityto distinguish between lexical and postlexical phonology in relation to clitics isargued for in Booij (1996b, 1997c).Phonologically conditioned selection of suppletive morphemes is

discussed in Carstairs (1988) The selection of allomorphs by output constraintsis argued for in Booij (1998), and Rubach and Booij (2001). The role ofparadigmatic relations in allomorphy is discussed in Booij (1997a, b).A good summary of the theory of Prosodic Morphology is McCarthy and

Prince (1998). Prosodic restrictions on affixation are discussed in Booij (2002b).Detailed analyses of reduplication are found in McCarthy and Prince (1998),Kager (1999), Wiltshire andMarantz (2000), and Struijke (2002).

184 interfaces

8Morphology and syntax:

demarcation and interaction

8.1 Words and phrases

There are at least four issues that have to be dealt with when we consider the

relation between morphology and syntax. One is the demarcation of the

empirical domains of these two modules of the grammar: when is a multi-

morphemic sequence a word, and when is it a phrase? Second, morphology

and syntax interact in two ways: syntactic constructs may form parts of

complex words, and syntax in its turn governs the use of morphological case

marking on words. A third domain of investigation is how morphological

operations may aVect the syntactic valency of words. Finally, languages may

have syntactic alternatives to the morphological expression of grammatical

and semantic content, and we might therefore want to know more about

the division of labour between morphology and syntax in this respect.

Let us Wrst have a closer look at the demarcation issue. Morphology

deals primarily with the structure of words, and syntax with the structure of

phrases. But how do you know if a particular combination of morphemes is

a word or a phrase? Is the lexical unit hard disk a word (that is, compound

of the type A þ N), or a noun phrase? How can we know?

8.1 Words and phrases 185

8.2 Grammatical functions

and case marking 190

8.3 Morphology and

syntactic valency 194

8.4 Periphrasis and

constructional idioms 201

Summary 203

Questions 204

Further reading 206

The most important criterion for wordhood is that of lexical integrity

that has already been discussed in section 4.2. The principle of Lexical

Integrity has been formulated as follows:

(1) ‘The syntax neither manipulates nor has access to the internal form of words’

(Anderson 1992: 84).

This principle implies that if we call something a word, it should exhibit

lexical integrity, that is, syntactic rules cannot refer to its parts. The

importance of this criterion can be illustrated by looking at the diVerence

between preWxed verbs and particle verbs in Dutch. This language has a

syntactic rule of Verb Second which requires Wnite verbal forms in main

clauses to appear in second position, after the Wrst constituent. If the verb

is preWxed, the preWx has to move along with the stem, which shows that

the preWx is not a separate word. On the other hand, if the verb is a

particle verb (that is a phrasal verb comparable to English word combin-

ations such as to look up), the particle is left behind. The underlying word

order in Dutch is SOV (Subject–Object–Verb). Hence, if the Wnite verb is

moved to second position, the particle is stranded in the original position

of the V. This shows that particles are words of their own. Hence, we get

the following contrast for the Dutch preWxed verb doorzoeken ‘‘to search’’

versus the particle verb doorzoeken ‘‘to continue searching’’:

(2) a. Jan door-zocht het hele gebouw

John through-seek.past.sg the whole building

‘‘John searched the whole building’’

b. Jan zocht tot 2 uur door

John seek.past.sg till 2 o’clock through

‘‘John continued searching till 2 o’clock’’

The diVerence between preWxed verbs and particle verbs is reXected by a

diVerence in the location of the main stress of these expressions: particles

bear main stress, whereas in preWxed verbs it is the verbal stem that carries

main stress.

Lexical integrity implies that English verb particle constructions such as

to look up are to be considered phrasal verbs because the two parts can be

separated as we saw in section 1.4:

(3) John looked up the information

John looked the information up

186 interfaces

Lexical integrity also manifests itself in the fact that syntactically gov-

erned rules of inXection do not apply to the individual parts of a word.

Consider the following names for kinds of cabbage in Dutch, both of the

form A þ N:

(4) rodeA koolN ‘‘red cabbage’’

zuurA koolN ‘‘lit. sour cabbage, sauerkraut’’

The Wrst cabbage name must be a phrase, witness the fact that the adjective

rod-e is inXected, as is the rule for Dutch adjectives in prenominal position.

Moreover, it carries phrasal stress, that is, main stress on the last constituent.

The second name is an AþN compound, not a phrase, because the adjective

zuur is not inXected. Main stress is on the Wrst constituent, as is the rule for

Dutch nominal compounds. The phrase zure kool is also possible, but

does not denote sauerkraut. Instead, it has the literal interpretation of

‘‘sour cabbage’’. Determining if an A þ N combination is a phrase or a

compound is more diYcult for English than for Dutch since English does not

inXect prenominal adjectives. In the case of English, our only formal criter-

ion is that of stress: in A þ N compounds main stress is usually on the Wrst

constituent and in A þ N phrases on the second, as in greenhouse versus

green house. On the basis of this criterion, wemay conclude that hard disk is a

compound since it has main stress on its Wrst constituent.

Lexical integrity pertains to syntactic rules. Phonological rules may have

access tomorphological structure, as do semantic rules. In the English phrase

a hard worker, for instance, the adjective hard functions as a modiWer of

the verbal base work in the noun worker. The phrase denotes someone who

works hard, not a worker who is hard. Thus, the rules of semantic interpret-

ation must have access to the internal structure of this deverbal noun.

It is a matter of debate to what extent rules of anaphora are subject to the

Lexical Integrity constraint. Consider the following sentences (Bosch 1983:

154):

(5) a. John likes [the guitar]i because he thinks iti is a social instrument

b. John became a guitarist because he thought it was a social instrument

c. Shakespearean imitators usually fail to capture his style

d. Fred is a New Yorker, but he has not lived there for years

In (5a) the pronoun it is interpreted as coreferential with the guitar. This is

indicated by the co-indexation of these two constituents by means of

a subscripted index i. In (5b), the pronoun it is also interpreted as the

morphology and syntax 187

guitar. Does this mean that the pronoun is co-indexed with a part of

the word guitarist ? If so, the Lexical Integrity constraint does not hold for

rules of anaphora. However, this goes against the general observation that

words, when embedded in complex words, lose their referential potential (in

fact, it is not words but phrases that refer to something). What is at stake in

sentence (5b) is that the pronoun it receives an interpretation within the

domain of discourse evoked by this sentence. The word guitarist that has a

transparent meaning will certainly evoke an entity ‘‘guitar ’’ in that discourse

domain. That is, the guitar is an inferred entity here, and the pronoun it

may be linked to that entity. The same applies to (5c, d) where Shakespeare

andNewYork are the inferred entities that function as the referents of his and

there respectively. What about the following sentence?

(6) John is an orphan, so he never knew them

Since orphan means ‘‘young person whose parents have died’’, we may

wonder if them in this sentence can refer to John’s parents? If that is the

case, we can conclude that the meaning of the simplex word orphan intro-

duced parents into the discourse domain. That is, the availability of particular

referents in a domain of discourse is primarily a matter of semantics. How-

ever, transparent morphological structure certainly helps to Wnd adequate

referents for pronouns in a discourse domain, and hence the sentences (5b–d)

are much better than sentence (6), to say the least.

The preceding remarks pertain to the issue of demarcation between

morphology and syntax: how can we determine if a particular multi-

morphemic construct is morphological or syntactic in nature? It is quite

clear that we need formal criteria. Semantic criteria such as semantic

idiosyncrasy are not of much help. The fact that yellow fever denotes

a speciWc disease is a semantic idiosyncrasy that shows that this morpheme

combination must be lexically stored. It is certainly a lexical unit, but not

necessarily a word in the morphological sense. Its stress pattern is that of

a phrase, with main stress on the head word: yellow fever.

A related question is whether phrases can be parts of words. Is there a

NoPhrase Constraint on complex words? The answer is negative: phrases do

occur as parts of words, as shown in (7) for English and Dutch compounds:

(7) a. [special exhibitions] gallery

[module for module] upgrade

[drugs and rehabilitation] centre

188 interfaces

b. [[oude mannen]NP [huis]N]N ‘‘old men’s home’’

[[gekke koeien]NP[ziekte]N]N ‘‘mad cow disease’’

[[hete lucht]NP[ballon]N]N ‘‘hot air balloon’’

[[God is dood]S [theologie]N]N ‘‘God-is-dead theology’’

In the Wrst example in (7a), the word sequence special exhibitions is a noun

phrase of the type A þ N. This is clear from its stress pattern (stress on

exhibitions) and the fact that the word exhibition is used in its plural form.

In the next two English examples we see the phrases module for module and

drugs and rehabilitation in the non-head position. The Dutch examples

show similar patterns, and the last example illustrates that even full

sentences can be embedded in a compound.

It is not the case that all kinds of phrases can feed word-formation. Noun

phrases with a determiner as parts of complex words are impossible in

English and Dutch:

(8) *the [[the special exhibitions] gallery]

*het [[de oude mannen] huis] ‘‘the old men’s home’’

This ungrammaticality has to do with the fact that the special exhibitions

and de oude mannen ‘‘the old men’’ with their deWnite determiners are

referring expressions, whereas word constituents in non-head position

have a modifying, classiWcatory function.

AYxes are not completely blocked from being attached to phrases, as

shown in (9):

(9) English: do good-er, do-it-your-self-er, fast-tracker, look-upper (from Ryder

2001)

Dutch: vierde klass-er ‘‘fourth grader, pupil of the fourth class’’, twaalfuur-tje

‘‘lit. twelve o’clock-dim, lunch food’’, onderons-je ‘‘lit. between us-dim, private

chat’’

AYxation of phrases is exceptional, however, and is restricted to very

productive aYxes such as the English nominalizing suYx -er, its Dutch

equivalent -er, and the very productive Dutch diminutive suYx -(t)je

which are the suYxes used in the examples in (9).

These examples of syntax feeding word-formation show that syntax and

morphology exhibit a speciWc form of interaction in that syntax can

feed word-formation. Of course, morphology also feeds syntax since

morphology provides units that are operated upon by syntactic rules.

You should realize that interaction is not the same thing as interface.

We reserve the term ‘interface’ for the ways in which diVerent kinds of


representations (phonological, formal, semantic) are linked to each other.

Here we looked at something diVerent, namely at constraints on the formal

structure of complex words, and more speciWcally whether pieces of syntax

can form parts of complex words.

A general observation concerning the interaction between morphology

and syntax is that they make use of the same word class categories:

morphological rules operate on words of a certain word class (noun,

verb, adjective, etc.), and also create words or word forms of a speciWc

category. There is thus a shared vocabulary for morphology and syntax

with respect to word classes.

8.2 Grammatical functions and case marking

In most languages, the interface between the semantic properties of a clause

and its morpho-syntactic structure (word order, case marking, etc.) is

partially regulated by the grammatical function frame of the verb of that

clause. This statement requires some explication, and we therefore begin by

analysing the grammatical properties of a simplex verb.

Consider the English verb to hit. This verb denotes an event in which at

least two participants are involved: the person who hits, and what is hit. We

refer to these two crucially involved entities as the core arguments of the

predicate HIT, and the Predicate Argument Structure (PAS) of this verb

can be represented as

(10) HIT, x, y

The PAS is an abstraction from the information contained in the Lexical-

Conceptual Structure (LCS) of a word. A complete LCS of to hit would

include that it denotes a movement in which an object is moved towards

another one, and makes contact with that object with considerable force.

This information is essential for a proper semantic interpretation, but we

focus here on those aspects that have direct relevance for the syntactic

behaviour of verbs, which are represented at the level of PAS.

The arguments of PAS are represented as variables such as x and y,

which will receive a particular value in each concrete sentence. In the

sentence John hit the ball, the value of x ¼ John, and the value of y ¼ the

ball. We can also add labels for particular semantic roles to these variables,

for instance Agent and Patient. The notion Agent refers to the entity that is

190 interfaces

in control of the event expressed by the verb. The Patient is involved in the

event, but is not in control. Such semantic roles (also called thematic roles)

are generalizations about the roles of participants in events denoted by

verbs. Our grammar of English now has to state that these two arguments

are normally expressed by the grammatical subject and the grammatical

object of the clause respectively. This relationship between the PAS of a

verb and its syntactic realization can be expressed as in (11) by linking the

grammatical functions ‘subject’ (SUBJ) and ‘object’ (OBJ) to the argu-

ments of PAS. The verb to hit is speciWed here as appearing with both a

subject and an object. This means that it is a transitive verb.

(11) HIT, xAGENT, yPATIENT Predicate Argument Structure

j jSUBJ OBJ grammatical function frame

The two arguments of to hit are the core arguments that always have to

be expressed. Most predicates require one or two arguments. There are also

verbs with three arguments, that have an indirect object expressing a third

argument, such as to give, as in John gave his sister a book in which his sister

has the semantic role of Goal (or Recipient). The number of arguments a

verb requires to be expressed is referred to as its syntactic valency.

In addition to the core arguments, there might be other entities involved

in specifying the event of hitting such as the instrument and the location, as

in John hit his enemy in the back with a stick. These latter speciWcations are

always optional, and such participants in the event are usually called

adjuncts. This enables us to reserve the term ‘argument’ for the core

arguments.

The relationship between the two levels of PAS and grammatical func-

tions is often predictable, by so-called linking rules. The following linking

rule is relevant for to hit. If there are two arguments, then the argument that

expresses the Agent of the action will be expressed as the grammatical

subject, and the other argument as the grammatical object. If there is

only one argument, the default linking rule applies: we have an intransitive

verb, and this argument will be linked to the grammatical function of

subject.

Word order is one of the means to indicate which particular grammatical

function is performed by a noun phrase. Linguists classify languages in

terms of the order in which grammatical functions are expressed: SVO


(Subject–Verb–Object), SOV (Subject–Object–Verb), etc. That is, we need

these grammatical functions for syntactic purposes. They also play a role in

a proper account of elliptical constructions, as in

(12) John hit his enemy, and (he) left

In this sentence, the subject of the second clause may be omitted. Here, we

interpret the person who left as John, not as his enemy. It is crucially the

subject position that can be gapped, and hence we must be able to refer to

grammatical functions such as Subject in our account of ellipsis.

In many languages, morphology is used to mark grammatical functions,

either through head marking or through dependent marking. Subject–

Verb agreement, for instance, marks the grammatical function Subject by

expressing some of the properties of the subject NP on the verbal head.

This marking may be used by the language user in order to identify the

subject of the clause. Case marking is a form of dependent marking which

signals the grammatical function of an NP in a clause.

In Indo-European languages with morphological case systems the dis-

tinction between grammatical subject and grammatical object is marked by

means of the opposition between nominative and accusative case. If there

is only one argument (the case of intransitive clauses), it is case-marked as

a nominative. When there are two arguments, the subject is marked

as nominative and the object as accusative. This system is called the

Nominative-Accusative system. An alternative case-marking system is

the Absolutive-Ergative system used in, among others, many Australian

languages. Usually, the symbols S, A, and O are used for the characteriza-

tion of these two systems (Dixon 1994: 6):

(13) S ¼ intransitive subject, A ¼ transitive subject, O ¼ transitive object

A stands for the subject of transitive sentences. It may be deWned as

follows: ‘[T]hat role which is likely to be most relevant to the success of

the activity will be identiWed as A’ (Dixon 1994: 7). S is the subject in

intransitive sentences, and O stands for the object in transitive sentences.

(Instead of O the symbol P (for Patient) is also used.) These two systems for

marking the grammatical functions can now be characterized as in (14). In

the Nominative-Accusative system A and S receive the same case marking,

whereas in the Absolutive-Ergative system this applies to O and S.

The following examples from German (15) and Dyirbal (16) illustrate the

two diVerent systems of case marking (Dixon 1994: 10).

192 interfaces

(14) PAS PREDICATE, x PREDICATE, x y

j j jGrammatical function S A O

j j jNominative-Accusative system nom nom acc

Absolutive-Ergative system abs erg abs

(15) Der Mann lach-t

the.masc.nom.sg man.masc.nom.sg laugh-pres.3sg

‘‘The man (S) laughs’’

Der Mann kauf-t ein

the.masc.nom.sg man.masc.nom.sg buy-pres.3sg. a.neut.sg.acc

Buch

book.neut.sg.acc

‘‘The man (A) buys a book (O)’’

(16) Numa banaga-nyu

father.abs return-nonfut

‘‘Father (S) returned’’

yabu Numa-Ngu bura-n

mother.abs father.erg see-nonfut

‘‘Father (A) saw mother (O)’’

In some cases it is not only the case marking that is ergative in nature, but

also the syntax, as can be seen in the construction of elliptical sentences. In

ellipsis, the second of two identical NPs in coordinated clauses is omitted,

as illustrated in (17) for English:

(17) John (A) saw his wife (O) and—(S) rejoiced (gapping of John under identity

of A and S)

In a language with ergative syntax such as Dyirbal, the subjects of intransi-

tive systems pattern with O, not with A:

(18) Numa banaga-nyu yabu-Ngu bura-n

father.abs return-nonfut mother-erg see-nonfut

‘‘Father (S) returned and mother (A) saw him (O)’’

There is no overt O in the second clause with the transitive verb bura ‘‘to

see’’, and the required O is taken to be identical to the S of the Wrst clause.

That is, the S of an intransitive clause and the O of a transitive clause

behave as instances of the same grammatical function.


Things can be even more complicated, due to the phenomenon of

split ergativity. In Dyirbal the morphological nominative-accusative

casemarking system is used for themorphologicalmarking ofWrst and second

person NPs; for all other NPs, the absolutive-ergative system is used. The

syntaxofDyirbal is consistently ergative,however, as shownby the fact that in

sentence (19), with nominative-accusative marking, it is the object of the Wrst

clause that controls the gapped subject in the second clause (Dixon 1994: 15):

(19) nyurra Nana-n bura-n banaga-nyu

you.all.nom we.all-acc see-nonfut return-nonfut

‘‘You all (A) saw us (O) and (we) (S) returned’’

In sentence (19) the omitted subject (S) of ‘‘returned’’ is interpreted as

‘‘we’’, that is, as being identical to the O of the preceding clause. This

shows that ergative morphology and ergative syntax are not the same, and

not necessarily linked: a language may have (partially) nominative-accusa-

tive morphology, but absolutive-ergative syntax.

8.3 Morphology and syntactic valency

The relation between the level of PAS and the level of grammatical func-

tions can be changed by morphological operations. The best known and

widely studied example of such a change is passivization. In this operation,

the Agent of the predicate is demoted to the status of an adjunct. The

remaining argument, if any, will then receive the status of S, and receive

nominative case in a nominative-accusative system. The following Green-

landic Eskimo sentences (from Fortescue 1984: 265) illustrate this process

for an ergative language:

(20) a. inuit nanuq taku-aat

people.erg.pl polar.bear.abs.sg see-3pl.3sg.ind

‘‘The people saw the polar bear’’

b. nanuq (inun-nit) taku-niquar-puq

polar.bear.abs.sg people-abl.pl see-pass-3sg.ind

‘‘The polar bear was seen (by the people)’’

In the second sentence, the passive suYx has the eVect of making the original

subject optional. It can be added as an adjunct, marked with ablative case,

whereas in the active sentence it is marked with ergative case. Note also that

194 interfaces

whereas in (20a) the person–number properties of both the plural Agent and

the singular Patient are marked on the verb, in (20b) it is only the properties

of the singular Patient that are marked on the verbal form. This is to be

expected given the non-argument status of the Agent in sentence (20b).

In present-day Romance and Germanic languages, there is no synthetic

passive form of verbs; instead, a periphrastic construction is used, consisting

of a passive auxiliary and a participle. In French, the verb etre ‘‘to be’’

functions as passive auxiliary, and theAgent can be expressed in a par-phrase:

(21) Je suis insult-e par Jean

I be.1sg.pres insult-past.ptcp.masc.sg by Jean

‘‘I am insulted by Jean’’

As stated above, passivization is an operation that aVects the mapping

between PAS and the grammatical function level. It is not an operation on the

semantic level (LCS or PAS) since it does not change meaning, only the form

inwhichmeaning is expressed. This is alsoprovenby the fact that thedemoted

Agent of the verbal predicate is semantically still available, for instance

as a controller. Notice the contrast in the following two English sentences:

(22) a. The boat was sunk to collect the insurance money

b. *The boat sank to collect the insurance money

In (22a), a passive sentence, the demoted Agent is still there semantically, as

the implicit subject of the verb collect in the embedded inWnitival clause. In

(22b), the intransitive, non-passive verb sink is used, and hence there is no

Agent involved at all. Consequently, an appropriate controller of the subject

of collect is not available in sentence (22b), and hence it is ungrammatical.

The essence of passivization is the demotion of the Agent argument, not

the promotion of the Patient argument. The promotion of the Patient to

the status of grammatical subject can be seen as the eVect of the default

linking rule: when there is only one argument, it will be expressed as the

grammatical subject. In conformity with this statement, Dutch and Ger-

man exhibit passivization of intransitive verbs, the so-called impersonal

passive, illustrated by the following example from Dutch:

(23) Er werd enthousiast gedanst

there aux.pass.past.impf enthusiastically dance.past.ptcp

‘‘There was enthusiastic dancing’’

In this sentence, the Agent is no longer mentioned explicitly, and since the

verb dansen ‘‘to dance’’ is intransitive, there is no other argument that can


be linked to the subject position. Since Dutch clauses with Wnite verbs

always require the presence of a subject, the subject position is Wlled with

a dummy word, the word er.

The diVerent patterns of linking arguments to grammatical functions are

also referred to as diVerent voices (active voice versus passive voice). Voices

express semantic relations between the subject and the action described by the

verb. Another example of voice is the middle voice. The grammarians of

Sanskrit and Greek speak of the middle voice as a distinct verbal form, just

as wemay have diVerent aspectual forms of a verb. In this case, the use of the

middle voice instead of the active voice does not mean that the mapping of

semantic roles on grammatical functions is diVerent, but the middle voice

marker changes the status of the subject with respect to the denoted action.

Here isanexample fromSanskrit (Klaiman1991:24, transcriptionsimpliWed):

(24) a. Devadattah katam karoti

Devadatta.nom mat.acc make.sg

‘‘Devadatta makes a mat’’

b. Devadattah katam krute

Devadatta-nom mat.acc make.sg

‘‘Devadatta makes himself a mat’’

As this example shows, the middle voice indicates that the subject is also the

beneWciary of the action expressed by the verb.

The term voice is also used for alternative markings of the pragmatic role

fulWlled by the noun phrases in a clause. The following example is from the

Mayan language Ixil (Ayres 1983, cited in Klaiman 1991: 34):

(25) a. A-k’oni in ta’n uula

2sg.erg-shoot 1sg.abs with sling

‘‘You shot me with a sling’’

b. Uula a-k’oni-b’e in

sling 2sg.erg-shoot-index 1sg.abs

‘‘With a sling you shot me’’

The index-suYx on the verb marks the promotion of the instrument to

sentence-initial position, with the instrumental preposition omitted. Thus,

focus is put on the instrument with which the shooting took place.

Instead of the Agent, the O can also be demoted. This is the anti-passive

construction, found in particular in languages with an absolutive-ergative

system. In this construction, the O can be omitted, or appear with

instrumental case. Hence, you get an intransitive sentence. An example

from Greenlandic Eskimo is the following (Fortescue 1984: 86):

196 interfaces

(26) a. inuit tuqup-pai

people.abs.pl kill-3sg.3pl.indic

‘‘He killed the people’’

b. inun-nik tuqut-si-vuq

people.instr kill-antipass-3sg.indic

‘‘He killed people’’

Sentence (26a) exhibits the normal case marking; in (26b) we have an

intransitive sentence, with the word for ‘‘people’’ marked with an instru-

mental case. The eVect of the anti-passive is to ‘despecify’ the direct object

of the transitive verb, if expressed at all. Hence, (26) does not refer to

speciWc people that were killed. Note also that the person–number proper-

ties of ‘‘people’’ are no longer marked on the verb.

Instead of demoting an argument, you may also promote it. This is done

in the applicative construction that is found in many African and Austro-

nesian languages. For instance, an instrumental or locative NP can be

promoted to the status of O as the eVect of adding an applicative aYx to

the verb, as in example (27) from the Atlantic language Wolof. In other

cases, an additional beneWciary argument is created, as in (28), from the

Mexican language Classical Nahuatl (Comrie 1985: 318):

(27) a. Mungi lekk ag kuddu

pres.3sg eat with spoon

‘‘He is eating with a spoon’’

b. Mungi lekk-e kuddu

pres.3sg eat-appl spoon

‘‘He is eating with a spoon’’

(28) Ni-c-no-paqui-lia

I-it-myself-wash-appl

‘‘I wash it for myself ’’

In sum, verbs may carry morphological markings that determine how

their arguments must be expressed on the level of syntactic structure. The

syntactic valency of verbs may thus be aVected by morphology.

Another way of changing transitive verbs into intransitive ones is noun

incorporation. What happens mostly in noun incorporation is that an

argument of the verb is not expressed by a separate NP, but as part of a

verbal compound. The verbal compound then functions as an intransitive

verb. The following example, a one-word-sentence, comes from the Amer-

indian language Tuscarora; the verbal compound of the form N þ V is

khw-

e

ti (Mithun 1999: 45):


(29) w-e-khw-

e

ti-?

factual-fem.agent-food-make-perf

‘‘she meal-made ¼ she cooked’’

In Tuscarora, verbs with incorporated nouns co-occur with verbs with

independent noun phrases. The diVerence is that the verb with incorpor-

ated noun denotes an institutionalized action, whereas the independent

noun phrases have referential potential. In the case of incorporation, the

argument usually receives a non-referential and non-individuated inter-

pretation. This functional diVerence is nicely illustrated by the fact that in

Tuscarora you can have constructions of the type She bread-made corn,

with both an incorporated noun and an independent NP, meaning ‘‘She

made corn-bread’’ (Mithun 1999: 46).

What is the precise nature of noun incorporation? Suppose we could coin

the verbal compound wood-chop in English (we can form the compound

wood-chopping, with a deverbal noun as head, but Germanic languages

tend to avoid verbal compounds). The verbal head chop is a transitive

predicate with two arguments. The non-head constituent wood will be

linked to the y-argument of chop. Put diVerently, the noun wood binds

the y-argument of chop, that is, it functions as the value of the y-variable.

So this argument cannot be expressed any more by an independent NP in

the relevant clause. Hence, the only argument left for independent syntactic

expression in a sentence is the x-argument. This implies that such verbal

compounds are intransitive. What we see here, then, is that the semantic

interpretation of a complex verb may lead to a syntactic valency that is

diVerent from that of its verbal head. Thus, unlike operations such as

passivization, noun incorporation does not aVect the interface between

syntax and semantics directly. The intransitivization eVect is the result of

the semantics of a particular word-formation process, in combination with

general principles concerning the relation between the Predicate Argument

Structure of a word and its syntactic expression.

The phenomenon of noun incorporation has led some linguists to

propose that some kinds of word-formation can be accounted for by

syntactic operations. In such a syntactic approach, the incorporated noun

is represented as an independent NP at the underlying syntactic level. The

noun is then adjoined to the V, resulting in a verbal compound. For

instance, the Wctive English verbal compound to bed-buy would be derived

as follows (with some simpliWcation of the syntactic structure). This is

represented in Figure 8.1.

198 interfaces

The noun bed originates in the object position of the VP, which is then

vacated; the empty position is indicated by t (for trace), and is co-indexed

with the moved N. Through this co-indexation, the incorporated N will be

interpreted semantically as the object of the verb. Thus, in this analysis

(defended by Baker 1988, 1996, for the Amerindian language Mohawk),

the intransitivization eVect of this kind of noun incorporation is the eVect

of a syntactic operation. The main challenge for this syntactic approach to

word-formation is providing a proper account of the semantic diVerences

between verbs with independent NPs and verbs with incorporated nouns

mentioned above.

The same debate as to the choice of a morphological or a syntactic

analysis of word-formation plays a role in the analysis of causatives.

Causative verbs are verbs in which the A has the role of the causer of an

event in which one or two entities play a role. The classic example of a

simplex causative verb in English is to kill, with the meaning ‘‘cause to die’’.

More precisely, the semantic structure of this predicate can be represented

as follows:

(30) CAUSE (x, (DIE, y))

This is a semantic representation of the fact that sentences with causative

verbs denote complex events, in this example the causation of the event of

someone dying.

Many languages have causative aYxes that turn non-causative verbs into

causative ones. The semantic eVect is the addition of a predicate CAUSE,

Fig. 8.1 Syntactic derivation of noun–verb compounds


and an additional argument, the causer. Hence, causativization has the

eVect of increasing the valency of words. This applies to adjectives, nouns

and verbs. The following examples are fromDiyari, an Australian language

(Austin 1981: 168), and from Turkish (Comrie 1985: 323), respectively.

(31) muka ‘‘sleep’’ (noun) > muka-Nanka ‘‘to put to sleep’’

kidi ‘‘clever’’ (adj.) > kidi-Nanka ‘‘to teach ¼ to make clever’’

(32) Ali Hasan-t ol-dur-du

Ali Hasan-do die-caus-past

‘‘Ali killed Hasan’’

mudur mektub-u imzala-dt

director letter-do sign-past

‘‘The director signed a letter’’

di�ci mektub-u mudur-e imzala-t-tt

dentist letter-do director-io sign-caus-past

‘‘The dentist made the director sign a letter’’

The last Turkish example shows how the addition of a causative suYx

increases the valency of a verb. Thus, it is an example of a valency-increasing

operation. We should realize, however, that causativization is a morpho-

logical process that primarily aVects the semantic properties of a predicate,

and hence the level of PAS. At the level of PAS, a CAUSER argument is

added to the PAS of the input word. This will then aVect its syntactic valency

through the linking rules that map arguments onto the grammatical function

frame of a verb. Therefore, although passive formation and causative forma-

tion both aVect syntactic valency, the nature of these operations is diVerent.

A syntactic approach to causativization assumes that the causative suYx

functions as a verb with a complement that denotes the caused event. The

syntactic structure of the Wrst Turkish sentence in (32) above might be

represented (in a somewhat simpliWed form and ignoring the past tense

suYx) as in Figure 8.2. This syntactic structure nicely expresses that a

sentence with a causative verb denotes a complex event. Since the causative

verb is a bound morpheme, it cannot surface as a word of its own, and must

be combined with a verbal stem. Hence, the verb of the embedded clause is

moved to the higher clause, and attached through adjunction to the left of

the causative suYx that functions as the verb of the main clause. This

change of structure should also aVect case assignment since the subject of

the embedded clauseHasanwill now have to be case-marked as object. This

kind of syntactic analysis of causatives is argued for in Baker (1988, 1996).

200 interfaces

8.4 Periphrasis and constructional idioms

The notion ‘periphrasis’ is used primarily in the analysis of inXectional

paradigms: for some cells of a paradigm, there is no speciWc morphological

form available. Instead, a word combination has to be used, that is, an

analytic form. Examples are the use of auxiliaries þ participles of main

verbs for the expression of perfect tense and the passive voice. In a number

of languages, progressive aspect is expressed by a periphrastic form of the

verb ‘to be’þ prepositional phrase, as illustrated by the following examples

from Dutch:

(33) Jan is [[aan]P [[het]Det [Wets-en]N]NP]PP

John is at the cycle-inf

‘‘John is cycling’’

(34) Jan is de aardappels aan het schillen

John is the potatoes at the peel-inf

‘‘John is peeling the potatoes’’

The unity of this aan het V-inf-construction as the periphrastic progressive

form of the verb is quite clear in sentence (34): the direct object de aardap-

pels ‘‘the potatoes’’ is not located right before the verb, as is normally the

case for objects in Dutch embedded clauses. Instead, the object precedes

the word sequence aan het ‘‘at the’’ that signals the progressive aspect.

Progressive aspect may also be expressed by using postural verbs such as

‘‘to sit’’ and ‘‘to stand’’ in coordination with a main verb, as illustrated here

for Afrikaans and for West-Flemish, a dialect of Dutch:

Fig. 8.2 A syntactic account of causative verbs


(35) Afrikaans

Piet staan ’n glas water en drink

Pete stands a glass water and drink

‘‘Pete is drinking a glass of water’’

West Flemish

Zij zat kousen en stoppen

she sat stockings and mend-inf

‘‘She was mending stockings’’

Note that it is only the postural verb in the West Flemish sentence that has

a Wnite form, the main verb appears in the inWnitive (in Afrikaans there is

no formal diVerence between inWnitive and Wnite forms).

Such constructions with a periphrastic function are constructional

idioms. Recall that constructional idioms are multi-word expressions that

are idiomatic in nature but not completely Wxed because some positions are

variable. For instance, we may speak of the Dutch constructional idiom

aan het V-inf, a sequence of word positions of which the Wrst two are Wxed,

but the third one is a variable position, which may be Wlled by the inWnitival

form of any verb that denotes an activity. Hence, this is a productive

pattern.

Many languages have preverb þ verb combinations that function simi-

larly to preWxed verbs. The notion preverb refers towords that appear before

verbs, and form a close unit with that verb. Quite often, these phrasal

constructions function as alternatives to preWxed verbs. This is the case

for particle verbs in Germanic languages, and preverbþ verb combinations

in Hungarian. As shown in section 8.1, particle verbs have a phrasal

status. The preverbs contribute to the aspectual properties of the verbal

predicate, and may also inXuence its syntactic valency. For instance, the

Dutch particle af expresses a result, and turns the intransitive verb werken

‘‘to work’’ into a resultative (hence telic and transitive) predicate, as in:

(36) Bettelou werkte haar opdrachten af

Bettelou worked her assignments particle

‘‘Bettelou Wnished her assignments’’

That is, Dutch has a constructional idiom af V with the meaning ‘‘to Wnish

V-ing’’ which may be speciWed in the lexicon. The meaning contribution of

the particle af is speciWed as part of the meaning of the constructional idiom

in which it occurs. The presence of the variableV indicates that new particle

verbs of this form can be formed. Particle verbs are therefore lexical units,

but not words in the morphological sense.

202 interfaces

Analytic causative constructions are also instantiations of constructional

idioms. In Germanic and Romance languages, the causative meaning is

often not expressed by an aYx, but by a separate causative verb such as

laten ‘‘to let’’ in Dutch and fare ‘‘to do’’ in Italian. The combination of the

causative and the main verb functions as a unit, and the recipient is marked

by a preposition, aan in Dutch, and a in Italian:

(37) Dutch

Ik liet het boek aan mijn collega zien

I let the book to my colleague see.inf

‘‘I showed the book to my colleague’’

Italian

Ho fatto vedere il libro a-l mio collega

have.1sg make-ptcp see.inf the book to-def my colleague

‘‘I showed the book to my colleague’’

The unitary nature of laat zien and ho fatto vederemanifests itself in the fact

that they select a recipient argument marked by a preposition, whereas

neither the causative verb nor the main verb select a recipient themselves.

Another type of constructional idiom that is functionally similar to

complex verbs is that of serial verbs, found in many African, Austronesian,

and Papua languages. The following examples are from Sao-Tomense, a

Portuguese-based creole spoken on the island of Sao-Tome in the Gulf of

Guinea. Characteristics of such constructions are that the two (or more)

verbs denote a single event, that there is only one overt subject, and one

tense marker (Hagemeijer 2001: 416):

(38) Bisu vwa subli

bird Xy.past go.up

‘‘The bird Xew upwards’’

Zon toma mantchin kota po

Zon take.past machete cut tree

‘‘Zon cut the tree with the machete’’

In sum, languages may have syntactic, analytic alternatives to the morpho-

logical expression of meaning. These syntactic alternatives may have the

status of lexical units and may exhibit special syntactic behaviour.

Summary

The relation between morphology and syntax must be dealt with from

a number of perspectives. One is the demarcation of the two: when is a


multi-morphemic sequence a word, and when is it a phrase? The criterion

of Lexical Integrity is the most important one for a proper delimitation of

morphology from syntax. Second, morphology and syntax interact in two

directions: syntactic constructs may form parts of complex words, and

syntax in its turn governs the use of morphological case marking on

words. The third perspective is that of syntactic valency: morphological

operations may aVect the syntactic valency of words. Finally, languages

may have analytic alternatives to the morphological expression of gram-

matical and semantic content. Productive types of word combination of

this kind can be qualiWed as constructional idioms.

Questions

1. Consider the following sentences of English:

a. John is a truck driver, and he often sleeps in itb. John is a truck driver, and an excellent one

What makes sentence (a) more difficult to interpret than sentence (b)?

2. English has the compound small claims court, with a phrasal left constituent.Why can we not use the phrase very small claims as the left constituent of acompound with court as its head (*very small claims court)?

3. Recently, a new English novel was advertised as unputdownable. Does thisword-formation imply that put down is one word?

4. The following sentence from Bolivian Quechua shows the use of the ac-cusative maker -ta (van de Kerke 1996: 26):

mikhu-chi-y wawa-taeat-CAUS-IMP child-ACC‘‘Make that the child eats’’

Explain why the word for ‘‘child’’ carries an accusative marking although it isthe Agent argument of eating.

5. Consider the following active sentence and its passive counterpart inTurkish (Kornfilt 1997: 324):

Hasan ders-ler-e basla-dI

Hasan lesson-PL-DAT begin-PAST‘‘Hasan began the lessons’’ders-ler-e basla-n-dI

lesson-PL-DAT begin-PASS-PAST‘‘The lessons were begun’’

a. What does this case marking pattern imply for the lexical specificationof the Turkish verb for to begin?

204 interfaces

b. Why does this pair of sentences support the claim that passive isdemotion of the Agent, and not promotion of the Patient?

6. Which kind of valency change can be observed by comparing the followingtwo Turkish sentences (Kornfilt 1997):

Hassan surahi-yi dolab-a koy-duHassan pitcher-ACC cupboard-DAT put-PAST‘‘Hassan put the pitcher into the cupboard’’Hassan-a surahi-yi dolab-a koy-dur-du-mHasan-DAT pitcher-ACC cupboard-DAT put-CAUS-PAST-1SG‘‘I made Hasan put the pitcher into the cupboard’’

7. Bolivian Quechua makes use of an assistive derivational morpheme -ysiwiththe meaning ‘‘to help’’, as illustrated by the following sentence (van deKerke 1996: 28):

mamma-y-ta wawqe-y-ta maylla-ysi-nimother-1SG-ACC brother-1SG-ACC wash-ASSISTIVE-1SG‘‘I help my mother to wash my brother’’

Explain why there are two nouns with accusative marking in this sentence.

8. Which kind of valency change is involved in the difference between the twofollowing sentences from the Austronesian language Kambera (Klamer1998: 201):

I Ama na-kei-yaj na menjajART father 3SG.NOM-buy-3SG.ACC ART table‘‘Father buys the table’’I Ama na-kei-nggak na menjajART father 3SG.NOM-buy-1SG.DAT ART table‘‘Father buys the table for me’’

(Identical subscripts indicate coreference of a pronominal marker on theverb and an NP.)

9. The following two Dutch passive sentences only differ in that the firstsentence has a present imperfective interpretation and the second a presentperfective one.

a. Jan word-t geslagenJohn become-PRES.3SG beat.PAST.PTCP‘‘John is beaten’’

b. Jan is geslagenJohn be.PRES.3SG beat.PAST.PTCP‘‘John has been beaten’’

Note that both finite verbs wordt and is are present imperfect tense forms ofthe verbs worden ‘‘to become’’ and zijn ‘‘to be’’ respectively. How mightthis unexpected interpretation of worden and zijn in combination with pastparticiples be accounted for in the grammar of Dutch?


10. Consider the following sentence from Middle Dutch:

Dit liep hi ende dede den coninc contThis walked he and did the king word‘‘He went and told this to the king’’

In this sentence, the object dit of the verbal expression dede cont ‘‘didword’’ precedes the intransitive verb liep ‘‘walked’’. How might thisextraordinary position of the object be explained?

Further reading

The issue of how to analyse the lexical conceptual structure of verbs is dis-cussed in Talmy (1985), and in a number of studies by Jackendoff (see Jackend-off 2002 for a synthesis). The relation between lexical semantics and argumentstructure is discussed in Dowty (1991). The distinction between LCS, PAS, andthe level of grammatical functions, and its relevance for morphology isdefended in Rappaport et al. (1993) and in Booij (1992). Valency change isdealt with in Dixon and Aikhenvald (2000). A typological survey of causatives ispresented in Song (1996).The language-independent criteria for categorizing an NP as a grammatical

subject are presented in Keenan (1976). Song (2001) gives a good survey ofcase assignment systems in the languages of the world. Dixon (1994) is adetailed study and survey of ergativity phenomena. There is an interestingdebate on the morphological versus the syntactic analysis of noun incorpor-ation. The syntactic approach is defended in Baker (1988, 1996), the morpho-logical approach in Mithun (1984, 1999).The notion ‘constructional idiom’ is discussed in Jackendoff (2002) and

Taylor (2002), its relevance for morphology is argued for in Booij (2002c).Preverbs are discussed in a number of articles in YoM 2003.

206 interfaces

9Morphology and semantics

9.1 Semantic interpretation of morphological structure

A basic idea of modern linguistics is that the relation between the meaning

and form of a simplex word is arbitrary. There is no particular reason why

a book should be denoted by the sound sequence [buk], as is the case for

English, and we might therefore say that the word book is an arbitrary

linguistic sign. If all linguistic signs were arbitrary, we would have to

memorize many linguistic expressions, and language would not be a very

Xexible communication system. Happily enough, the arbitrariness in the

form–meaning relation of linguistic expressions is reduced by their having a

layered structure: sentences are not holistic signs, and can be divided into

constituents and ultimately words (syntactic structure), and words in their

turn may have internal structure themselves (complex words). Language is

clearly a combinatorial system. The meaning of complex words is not

completely arbitrary, but (at least partially) motivated. It is an obvious

task for morphologists to investigate the regularities involved in assigning

a particular meaning to a complex word.

The most general principle for the semantic interpretation of both

morphological and syntactic structure is the Compositionality Principle:

(1) The meaning of a complex expression is a compositional function of that of its

constituents, and the way they are combined.

9.1 Semantic interpretation of

morphological structure 207

9.2 Semantics and syntactic

valency 214

9.3 Polysemy 220

Summary 225

Questions 226

Further reading 227

For morphology, this implies that we can derive the meaning of a complex

word on the basis of its internal structure.

Compositionality is an important, but fairly general principle, and the

content of this compositional function requires more detailed speciWcation.

A Wrst speciWcation of this compositional function is that semantic scope

reXects structural hierarchy. Consider the two following nouns from the

Amerindian language Yup’ik (Mithun 1999: 43):

(2) [yug-pag]-cuar [yug-cuar]-pag

person-big-little person-little-big

‘‘little giant’’ ‘‘big midget’’

In the example on the left, the suYx -cuar ‘‘little’’ has been added to the

stem after the suYx -pag ‘‘big’’. Hence, -cuar has semantic scope over yug-

pag. This means that the meaning of -cuar does not modify the meaning of

yug only, but the meaning of the whole stem yug-pag. In the right example,

the inverse situation applies. Hence, there are semantic diVerences between

these two words with the same set of morphemes which reXect diVerences in

morphological structure. A similar observation can be made about the

English adjective unbelievable. This adjective has the meaning ‘‘cannot be

believed’’, and not the meaning ‘‘can be not believed’’. This reXects the fact

that the structure of this adjective is as follows:

(3) [un[[believ]Vable]A]A

That is, the preWx un- has scope over the stem believable.

In the case of compounds, it is the notion ‘head’ that we need for

a proper semantic interpretation: the meaning of the non-head of a

compound functions as a modiWer of that of the head. This latter statement

is a correspondence rule that speciWes the interface between the formal

structure of a compound and its semantic interpretation. Consider the

following nominal compounds of Dutch:

(4) graan-molen ‘‘corn mill’’

mosterd-molen ‘‘mustard mill’’

verf-molen ‘‘paint mill’’

water-molen ‘‘water mill’’

wind-molen ‘‘wind mill’’

TheseWve compounds all denote a particular kindofmolen ‘‘mill’’. This shows

that the structural notion ‘head of a compound’ that we need for formal

208 interfaces

reasons (such as the fact that the head determines the gender of the whole

compound) is essential for the semantic interpretation of compounds as well.

Headship induces the ‘is a’ relation: a graanmolen is amolen. The relationship

between head and non-head can be circumscribed as ‘has some relation to’.

This is an intentionally vague semantic qualiWcation. When we paraphrase

theseWvecompoundswegetquitediVerentcircumscriptionsof theirmeanings:

(5) graan-molen: mill that grinds corn

mosterd-molen: mill that grinds mustard seeds for the production of mustard

verf-molen: mill that grinds wood for the production of paint

water-molen: (i) mill that is powered by water, (ii) mill that transports water

wind-molen: mill that is powered by wind

The semantic paraphrase ‘has some relation to’ can be referred to as R.What

we have to say for languages with right-headed compounds is that the

compound structure AB correlates with the semantic structure R(B,A), that

is, ‘B has some relation to A’. The exact nature of this relation is not a matter

of the grammar, but of world knowledge (also called encyclopedic knowledge)

and, in some cases, knowledge of the context in which a compound is used.

The compounds in (4–5) are all established compounds, whichmeans that the

speciWc interpretation of R has been Wxed. For new compounds, the content

of R can be inferred at the time of its being uttered on the basis of knowledge

of the world and/or context. For instance, when I came across the English

compound umbrella organization for the Wrst time, I had no problem in

understanding it, and I did not have to look it up in my English dictionary.

The interpretation is obviously ‘‘organization that functions like an

umbrella’’. An important semantic component of words is their function,

and in this example umbrella is a further speciWcation of that function. In

addition, this compound shows that language users are able to interpret

metaphorical use of language. In metaphors, notions from one domain of

knowledge are transferred to another domain.We know that an umbrella has

a protective function, and can keep one or more persons protected against

rain or sun under its screen. This knowledge is then transferred to the more

abstract domain of organizational structure, leading to the interpretation

that we have to do here with some organizational superstructure.

It is useful to distinguish between the notionsmeaning and interpretation.

The meaning contribution of right-headed compound structures of the type

AB is R(B,A), and this relationship then receives a speciWc interpretation

for each individual compound, by means of interpretational mechanisms as

morphology and semantics 209

those discussed above. These mechanisms are pragmatic in nature since

they follow from the pragmatic principle of cooperation between speaker

and hearer: try to come up with the most sensible interpretation possible so

that we understand each other.

The high degree of abstractness (or vagueness) of the meaning contribu-

tion of compound structure makes compounding an extremely Xexible

device from a semantic point of view. This, in combination with its trans-

parent morphological structure, undoubtedly contributes to its enormous

productivity in many languages.

A similar proWtable vagueness can be observed in two other wordforma-

tion processes, conversion and the derivation of relational adjectives.

Consider the following cases of conversion in Dutch:

(6) Noun Verb

bier ‘‘beer’’ bier ‘‘drink beer’’

kaas ‘‘cheese’’ kaas ‘‘produce cheese’’

melk ‘‘milk’’ melk ‘‘take milk from an animal’’

tafel ‘‘table’’ tafel ‘‘dine’’

water ‘‘urine’’ water ‘‘urinate’’

These verbs all have the meaning ‘‘action that has some relationship R with

the base-noun’’, but the exact nature of this relation varies from word to

word. Clearly, the meaning of these words is compositionally determinable,

but we need other, non-linguistic resources to come up with a speciWc inter-

pretation of R. Once they have been coined, with a speciWc interpretation,

that interpretation might become the conventionalized interpretation, often

registered in dictionaries. Storage of the conventionalized interpretation is

particularly a must for English verbs derived from proper nouns such as to

boycott, to hoover, and to xerox. The proper noun Boycott as such does not

give you a clue as to what a verb to boycottmight mean. Hence, the meaning

of these verbs is not computable on the basis of the meaning of their base

nouns without further information on these proper names.

Abstractness of the meaning contribution of morphological structure is

also the key to understanding the manifold interpretations of reduplicative

structures such as plurality, iterativity, and intensity. Consider the follow-

ing sentences from Afrikaans (Botha 1988: 92–3):

(7) a. Die kinders drink bottels-bottels limonade

the children drink bottles-bottles lemonade

‘‘The children drink many bottles of lemonade’’

210 interfaces

b. Hij lek-lek oor sy droe lippe

he lick-lick over his dry lips

‘‘He licks and relicks his dry lips’’

The following rule accounts for the diVerent interpretations of these

reduplicative structures:

(8) Interpret ai ai as [A increased] (where A represents the meaning of a and

increased represents an abstract semantic unit) (Botha 1988: 94).

This is a correspondence rule that relates a particular morphological form to

its semantic structure. The abstract meaning increased will then receive a

more speciWc interpretation by means of conceptualization rules which deWne

conceptual well-formedness. For instance, the following conceptualization

rulemay be assumed: if A has themeaning component countable thing, the

property increased must receive a numerical interpretation. This is the case

in sentence (7a). In example (7b), it is a verb that is reduplicated. Verbs denote

events with the measurable property of duration. Hence, with verbs the

notion increased should be interpreted as increase of time, another example

of a conceptualization rule. However, licking one’s lips is a bounded event.

Hence, the only sensible interpretation here is an iterative one: repeated

licking must be involved. This rule of iterative interpretation of bounded

events is a third example of a conceptualization rule. Such rules thus do not

specify the interface between formal structure and semantic interpretation,

but further develop the semantic interpretation.

This analysis of the interpretation of complex words shows that it is a

dynamic and Xexible process. This is also the case for the interpretation

of complex adjectives. Many kinds of denominal adjectives function as

relational adjectives (cf. section 4.1). Relational adjectives denote the

existence of a relation between the noun that they modify and some other

entity evoked by that adjective. They are distinguished from qualifying

adjectives that denote a quality of the noun they modify. Consider the

following examples from English:

(9) the Americ-an Xag

a person-al computer

a spous-al hire

The meaning contributions of the diVerent denominal suYxes involved are

all the same: ‘‘related to base noun’’. Thus, they relate the base noun of

the adjective to the head noun that these adjectives modify in a phrase. An


American Xag does not denote a Xag that is American in nature, but the Xag

of America. Due to this relational nature of such adjectives, they cannot be

used in predicative position (except when contrast is involved, as in (10b),

nor can they be modiWed, unlike qualitative adjectives (10c):

(10) a. *The Xag is American/*a very American Xag

b. That Xag is not American but Canadian

c. That Xag is blue/a very blue Xag

The same applies to the other two examples in (9): a personal computer is a

computer meant for use by individual persons, not a computer that is

personal in nature. A spousal hire denotes the situation in which the

employer not only hires a person, but also her or his spouse.

By adding a modiWer, relational adjectives can be forced to be inter-

preted as qualifying adjectives. If I call someone a very American lady,

I invoke all the prototypical qualities of American ladies (wearing shorts in

summer, speaking loudly and with a lot of gestures, etc.). This kind of

interpretational shift is thus a case of type coercion: modiWers require

qualifying adjectives as their adjectival heads, and hence, in such cases

denominal adjectives are interpreted as qualifying adjectives. This is

another demonstration of the dynamics and Xexibility of meaning assign-

ment to complex words.

Relational adjectives play an important though not exclusive role in

what have been called bracketing paradoxes. For instance, a moral philoso-

pher is not a philosopher who is moral, but someone who deals with moral

philosophy. Hence, there seems to be a mismatch between the formal

structure of this phrase, and its semantic interpretation:

(11) syntax: [[moral]A [philosopher]N]NP

semantics: [[moral philosophy]er]

The same observation applies to phrases such as nuclear physicist, criminal

lawyer, small farmer, and Wrst violinist. On second thought, however, there

is no bracketing paradox involved. The same kind of interpretation occurs

when there is no possibility of bracketing the phrase diVerently at the

semantic level, as shown by the following examples with underived nouns:

(12) a good athlete, an old friend

A good athlete is not necessarily a good person, but someone who is good as

an athlete. Similarly, an old friend is not necessarily old, but someone the

212 interfaces

friendship with whom is old. In these cases, this interpretation cannot be

attributed to a diVerence in bracketing between the formal and the seman-

tic structure. What we need therefore is a semantic principle that tells us

how to interpret such phrases, both those with denominal relational adjec-

tives and those with simplex adjectives. The general idea is that an attribu-

tive adjective, whether a qualifying adjective or a relational adjective, may

modify only part of the semantic structure of the head noun. The phrase old

friend can mean ‘‘a friend who is old’’. In that case, old only says something

of what we may call the person component of the meaning of friend. If we

interpret this phrase as ‘‘someone who has been a friend for a long time’’,

the adjective mentions a property of another meaning component, the

function component ‘‘friendship’’. Denominal relational adjectives have

the speciWc property that the entity invoked by their base noun, for instance

crime in the case of criminal, functions as an argument of the function

component of the meaning of lawyer. This functionmay be circumscribed

as ‘‘giving advice on legal matters concerning x’’. In the phrase criminal

lawyer the x will then be taken to stand for ‘‘crime’’.

How does the interpretation of morphological structure proceed in those

cases in which the morphological operation involved does not consist of

concatenation, but of operations such as vowel change? How do we deal

with the interpretation of the past tense form saw of to see, in which there

is no separate past tense morpheme? A possible answer is that such oper-

ations are triggered by the presence of morphosyntactic features such as

[þpast] or an abstract grammatical morpheme past. Such features or

morphemes will then be linked to a semantic property past at the semantic

level. The linking rules will also specify the scope of past. The semantic

scope of this property is not just the verb itself, but the whole event

described, as represented in a semi-formal way in (13):

(13) Indriaas saw the accident , past [see, Indriaas, the accident]

Thus, it is speciWed that the event of Indriaas seeing the accident took place

before the moment of speaking.

In sum, the Compositionality Principle is the main principle of interface

between formal (morphological and syntactic) structure and semantic

structure. This general principle requires further speciWcations of the sort

discussed above, in order to do justice to the complexities of this interface.

In addition, there are conceptualization rules pertaining to the semantic

level only that further enrich semantic structure.


9.2 Semantics and syntactic valency

The relation between morphology and syntactic valency has already been

broached in section 8.3. However, it also needs attention in a chapter on

morphology and semantics because the syntactic valency of a complex

word is (at least partially) determined by its semantic properties.

As observed in section 1.3, one of the functions of morphology is to

change the syntactic category of words with the eVect that other syntactic

uses are made possible. A cross-linguistically very common form is that of

action nominalization, in which verbal constructions are deverbalized, and

acquire a noun-like behaviour. There are degrees of deverbalization as

the following examples of deverbal nominalization in English show

(Malouf 2000: 93):

(14) a. Chris was shocked that Pat illegally destroyed the evidence

b. Chris was shocked by Pat having illegally destroyed the evidence

c. Chris was shocked by Pat’s having illegally destroyed the evidence

d. Chris was shocked by Pat’s illegal destroying of the evidence

e. Chris was shocked by Pat’s illegal destruction of the evidence

In (14b–d) we see the gerundive nominals having destroyed and destroying,

and in (14e) the deverbal action noun destruction. In the sentences (14b–e),

these deverbal nouns are used as the heads of a noun phrase. This can be

concluded from the fact that they occur in a PP with the preposition by.

Yet, they show some verb-like behaviour as well. This is most clearly the

case with the gerundive nominals that co-occur with the adverb illegally,

which is impossible to use with the derived noun destruction. Selecting

adverbial forms as modiWers is a distinguishing property of verbs and

adjectives, whereas nouns select adjectives as modiWers. Furthermore, in

(14b, c), the object of the verb destroy, the phrase the evidence, appears as a

bare NP, without a preposition. This is of course a prototypical property of

verbs rather than of nouns. In (14e), the complement-NP of destruction has

to be preceded by a preposition, the default preposition of.

It will be clear then that such nominalization patterns pose quite an

analytical challenge to linguists, and raise the issue whether categories such

as V and N can always be sharply distinguished. This is why deverbal

nominalizations have been qualiWed as belonging to the set of mixed

categories (also called transcategorial constructions). Other examples of

such mixed categories are verbal participles. As we saw in section 5.1 for

214 interfaces

German, they behave as adjectives with respect to inXection, and as verbs

with respect to the kind of phrases with which they can combine.

These observations suggest that the function of morphology might

be that of category change only, without any particular semantic eVect.

The (partial) preservation of the syntactic valency of the base of such

complex words is referred to as inheritance. Deverbal nouns in -ion diVer

from gerundive nominals in exhibiting a lesser degree of inheritance of

verbal properties, as we saw above. In deverbal nominalization of this type,

the deverbal noun inherits the Predicate Argument Structure of its base

verb, but the syntactic expression of the arguments is diVerent from that of

the base verb. They must be expressed in an of-PP, or in the prenominal

possessive form with ’s. Moreover, the syntactic expression of the inherited

arguments of such deverbal nouns seems to be optional; compare:

(15) a. Pat’s destruction of the evidence

b. the destruction of the evidence

c. Pat’s destruction

d. the destruction

In (15c), Pat is interpreted as the Patient of destruction, not as Agent, unless

there is some speciWc context in which objects can be omitted.

Action nouns can further develop fromsimple event nouns into result nouns

denoting the result of the action. In the latter interpretation they no longer

denote an event with participants, and hence do not allow for Agent phrases.

When such nouns are interpreted as result nouns, they can be pluralized.

Therefore, such plural deverbal nouns exclude the use of Agent phrases:

(16) Extensive collections of shells (*by Indriaas)

These expressions (*by my father) are old-fashioned

In sum, when there are diVerences in syntactic valency between verbs and

their corresponding deverbal nouns, these diVerences can be seen as reXect-

ing diVerences in their semantic structure. This shows once more the strong

dependence of syntactic valency on the semantic properties of words.

The issue of inheritance of syntactic valency also crops up in the analysis

of compounds of the following type:

(17) coVee-maker, pasta-eater, story-teller, truck-driver

In these compounds headed by a deverbal agent noun, the left constituent

is interpreted preferably as the Patient of the action denoted by the


verbal base of the head noun. How are we going to account for that

interpretation? We cannot consider these words to be derived by means

of the suYx -er from verbal compounds of the N þ V type, since this is not

a productive type of compounding in English: compounds such as to

coVee-make or to pasta-eat cannot be formed. Hence, a word such as

coVee-maker must be considered a case of N þ N compounding, with the

head noun being a deverbal noun.

There are two ways of accounting for the Patient interpretation of the left

constituent. The default one is to simply assume that the Patient interpret-

ation is a Wlling-in of the general relationR that exists between the two parts

of a compound. For maker, the most sensible interpretation of the relation

with coVee is that coVee is a speciWcation of what is made. The other

analytical option is to assume that deverbal nouns keep the Predicate

Argument Structure of their verbal base. The semantic structure of maker

will then be as follows, with the PAS of its verbal base incorporated:

(18) x [make, xAGENT, yPATIENT]

This semantic structure is a slightly more formal representation of the

informal description of the meaning of maker as ‘‘one who makes some-

thing’’. The semantic eVect of the -er-suYx is that it binds the x-argument of

the verbal base. This means that this argument cannot be expressed any

more. Hence, only the y-argument is left to be expressed by a nominal,

either as the left constituent of a compound, or as an NP that is preceded by

the default preposition of. Both the compound coVee-maker and the phrase

maker of coVee are well-formed expressions.

A possible argument for the second analytical option based on facts of

Dutch is that this language has a number of compounds of which the

deverbal head only occurs in compounds or with a PP-complement

(Booij 1988):

(19) a. ijs-bereider/bereider van ijs ‘‘ice maker/maker of ice’’

b. bevel-hebber ‘‘lit. command haver, commander’’

c. woord-voerder ‘‘lit. word carrier, spokesman’’

The nouns bereider and voerder do not occur outside these contexts, and the

deverbal noun hebber does occur as a noun by itself, but only in the idio-

syncratic interpretation ‘‘greedy person’’. The verbs involved (bereiden ‘‘to

prepare’’, hebben ‘‘to have’’, and voeren ‘‘to carry’’ are obligatorily transitive

verbs, with two arguments. If argument structure is inherited by deverbal

216 interfaces

nouns in -er, it is predicted that the second argument must be expressed,

either morphologically or syntactically, as in (19a). The verb hebben

‘‘to have’’ is also obligatorily transitive. In the lexicalized compound

bevelhebber the x-argument is bound by -er, and the y-argument by

bevel ‘‘command’’. There is an alternative analysis, however, for words such

as bevelhebber andwoordvoerder. Theymight have been derived directly from

the Dutch lexical units bevel hebben ‘‘to command’’ and woord voeren ‘‘to

speak’’. In that case, we do need the nouns hebber and voerder as building

blocks of these words. In short, there is no established analysis of such facts.

The inheritance of syntactic valency as analysed here is the eVect of

preservation of argument structure of the base words. There is, however,

also a kind of inheritance of a more syntactic nature, with little or

no semantics involved. Verbs may select a speciWc preposition for the

complement they occur with, a prepositional object. This also applies to a

variety of adjectives. This choice of preposition is an idiosyncrasy that has

to be encoded in the lexicon, and this choice is taken over by the derived

word, as illustrated in (20):

(20) to long for longing for

to compare to comparable to

to hope for hope for

to trust in trust in

divide by divisible by

averse to averseness to

curious about curiosity about

susceptible to susceptibility to

This implies that in these cases there is a transfer of speciWc syntactic

subcategorization information from the base word to its derivative.

The relation between the semantics of a word-formation process and its

eVect on syntactic valency is also at stake in two other English word-

formation processes, the coining of deverbal adjectives ending in -able,

and middle verbs:

(21) do-able ‘‘can be done’’

drink-able ‘‘can be drunk, Wt for drinking’’

read-able ‘‘can be read, pleasant to read’’

wash-able ‘‘can be washed’’

(22) This book reads easily

These products sell well

This chicken broils excellently


Both processes serve to create words that do not denote events but stative

properties. In this respect they are quite unlike the passive construction,

which does denote events. Since these types of word do not denote events,

they cannot occur with an Agent phrase. The only argument that is

expressed is the Patient-argument of the base verb. This argument is stated

to have the property that it can participate in the kind of event denoted by

the base verb. Thus, sentences like the following are ungrammatical:

(23) *This tea is drinkable by Mary

*This book reads easily by John

The semantic structure of these adjectives and verbs implies that there is no

Agent to express, not even in a by-phrase, because there is no Agent-

participant in the Lexical Conceptual Structure of these words, and hence

they are intransitive predicates.

You may have noticed that my gloss of drinkable is somewhat imprecise.

If we say of a wine that it is drinkable, we mean that it is a good wine to

drink.Whatwe observe here, is amore general pragmaticNon-Redundancy

Constraint involved in the interpretation of words. It seems superXuous to

say of wine that is drinkable, meaning that it can be drunk. Of course it can!

Therefore, if someone nevertheless states that it is drinkable, we infer that

this must mean it is good wine to drink, otherwise the information would be

superXuous. We assume that speakers try to be informative, after all, and

to provide relevant information. For the same reason, it seems odd to

say that a book can be read. Thus, adjectives such as drinkable and readable

receive an interpretation of positive evaluation. The same holds for

middle verbs. In their concrete use in sentences, middle verbs are always

accompanied by an evaluative adverb such as well or easily.

This pragmatic principle may also explain why denominal adjectives in

-ed such as eyed do not occur on their own, but in compounds only:

(24) ?eyed blue-eyed

?faced red-faced

?Wsted two-Wsted

?handed two-handed

?handed right-handed

The adjectives in the left column of (24) are morphologically well-formed.

However, they violate the pragmatic Non-Redundancy Constraint. It is

expected that human beings have a face, and hence it does not seem to

make much sense from a pragmatic point of view to say My husband is

faced. Hence the use of such adjectives is infelicitous (indicated here by the

218 interfaces

question mark) unless they are embedded in a compound. Yet, this might

not be the whole story. If a man has a lot of hair we prefer to call him a

hairy man, and a haired man sounds odd although it makes sense from the

pragmatic point of view.

The crucial role of the semantics in determining the syntactic valency of

derived verbs can be illustrated by the formation of Dutch verbs by means

of the preWx be-. Morphology may be used to form obligatorily transitive

verbs from verbs that are intransitive or optionally transitive. In Dutch, the

preWx be- is used for this purpose as shown in (25). These examples show

that, whatever the syntactic valency of the input verb, the syntactic valency

of the output verb is always that of an obligatorily transitive verb. The

Lexical Conceptual Structure of deverbal be-verbs can be circumscribed as

follows: ‘x completely aVects y by executing the action expressed by the

base verb’. This means that the Predicate Argument Structure of such verbs

always comprises two arguments. Moreover, the resulting verbs are always

telic verbs because the endpoint of the action is implied: the y-argument is

completely aVected, and thus the end of the action is implied.

(25) Type of input verb Output verb

intransitive

loop ‘‘to walk’’ iets beloop ‘‘to walk on

something’’

klim ‘‘to climb’’ iets beklim ‘‘to climb on

with prepositional object something’’

aan iets twijfel ‘‘to doubt about

something’’

iets betwijfel ‘‘to doubt

something’’

met iemand vecht ‘‘to Wght with

somebody’’

iemand bevecht ‘‘to Wght

somebody’’

with direct object

iets schilder ‘‘to paint something’’ iets beschilder ‘‘to cover

something with paint’’

iets plak ‘‘to glue something’’ iets beplak (met iets) ‘‘to cover

something (with something)’’

with direct and prepositional object

iets van iemand roof ‘‘to rob some-

one from something’’

iemand beroof (van iets) ‘‘to rob

somebody (from something)’’

iets in iets plant ‘‘to plant something

in something’’

iers beplant (met iets) ‘‘to plant

something (with something)’’

Source: Booij 2002a: 192.


We can now predict the syntactic valency of be-verbs on the basis of the

following rule: ‘if a telic verb has a Patient-argument, this argument

must always be expressed syntactically’. This regularity is an example of a

correspondence rule that speciWes the interfacebetween syntax and semantics.

Thus, the fact that thesebe-verbs areobligatorily transitive is fullypredictable.

In sum, the eVects of morphological operations on the syntactic valency of

input words are primarily the eVects of the semantic changes brought about

by these operations. If the semantic changes are minimal, the diVerences in

syntactic valency between base word and derived word will also be minimal.

9.3 Polysemy

Consider the following list of English deverbal nouns in -er, subdivided into

a number of semantic categories:

(26) Agent baker, writer

Instrument mower, pointer

Experiencer feeler, hearer

Action breather, disclaimer

Locative diner, sleeper

Clearly, it is not the case that the suYx -er is used only to derive agent

nouns. So should we assume a number of diVerent suYxes -er, each with a

meaning of their own? Well, there is certainly more than one suYx -er in

English: there is also the comparative suYx -er, with a completely diVerent

meaning. That is, there are at least two homophonous suYxes -er. In order

to interpret the variation in meaning for the deverbal suYx -er, another

notion is more appropriate, that of polysemy. We speak of polysemy

when a morpheme or a word has more than one meaning, but with

some systematic relation between these diVerent meanings. Consider the

following meanings for the word head:

(27) 1. The top part of the body in humans and some other higher animals;

2. Person (as in dinner at 20 dollars a head );

3. Leader

Actually, there are many more meanings for head, but these three suYce for

our purposes. Meaning 1 is the primary meaning of this word. Meaning 2

220 interfaces

can be considered to be derived from this primary meaning through the

semantic mechanism of metonymy in which a word is interpreted as denot-

ing something that is associated with the object that it literally denotes. In

this example, the word head is used to denote the person who is the owner

of the head. In the third meaning, we speak of metaphorical use: just as a

head in its literal sense governs the body to which it belongs, a human being

may govern an institution of some sort (which can also be called a ‘body’).

Thus, these three diVerent meanings of head can be related systematically in

terms of general mechanisms of semantic interpretation, and we consider

this a case of polysemy.

There are also systematic explanations for what one might call the poly-

semy of the suYx -er. To begin with, the subject argument of the underlying

verb may be a human agent (baker), a non-human, impersonal one (pointer

‘‘a rod used to direct attention’’) or an experiencer (hearer). Hence,

the diVerent interpretations correlate with the diVerent semantic roles of

the subject arguments of the base verbs. In the case of sleeper ‘‘sleeping car’’

something else is at stake. Here, the location (or space) of the action of

sleeping is denoted. This interpretation can be understood by means of the

notion domain shift: one may go from one semantic domain to another,

related one, and thus derive new interpretations. An example of such a

cross-linguistically valid domain shift chain is:

(28) PERSON>OBJECT>ACTIVITY>SPACE>TIME>QUALITY (Heine

et al. 1991: 48)

In this chain, one might go from left to right, from one domain to the next.

For instance, the English word reader, which is a word for a person, can also

be shifted to the domain ofOBJECT, and be interpreted as a kind of patient:

‘‘collection of reading material’’. Similarly, the word diner that can denote a

person who dines can also denote a space where one dines, a (speciWc

American type of) restaurant. The possibility of such domain shifts also

explains the activity name disclaimer and the space or location name sleeper.

These kinds of domain shifts are examples of metaphorical and meto-

nymical extensions. Metaphorical operations conceptualize domains of

cognition in terms of other, usually less abstract domains. The shift from

PERSON to INSTRUMENT is an example of domain shift that one often

Wnds in natural languages. This chain can be seen as a metaphorical one:

the notion AGENT is transferred to the domain of inanimate material

things that are conceived of as agents that perform a particular task.


A locative interpretation (as in diner) looks more like metonymy: the

name for the agent is transferred to the location of the action denoted by

the base verb.

In (29) deverbal nouns of Hungarian (Comrie and Thompson 1985: 355)

and Dutch (Booij 1986) illustrate that these domain shifts occur cross-

linguistically. This is what should be the case since such interpretational

chains are cognitive in nature, and hence applicable to diVerent languages.

Note, moreover, that quite a number of such nouns have more than one

interpretation. The Dutch agent noun sender can be interpreted as an agent

noun denoting a human being, but also as a radio or tv channel, or an

instrument for transmitting signals. If the instrument interpretation is the

established one, as is the case forDutch opener (an instrument to open cans or

bottles), the (human) agent interpretation can always be activated and used,

as in De opener van deze expositie ‘‘The opener of this exhibition’’. The

American English word diner for a certain type of restaurant can still be used

for someone who dines. This means that we do not have to assume a list of

diVerent meanings for these deverbal suYxes. It is the individual complex

words with these suYxes that show this semantic variation, and this variation

follows from independently established principles of semantic interpretation.

(29) Hungarian (deverbal suYx -o/o)

Agent ır-o ‘‘writer’’

Instrument hegyez-o ‘‘pencil sharpener’’

Location tarsalg-o ‘‘parlor’’ < tarsalg ‘‘to talk’’

Dutch (deverbal suYx -er)

Agent schrijv-er ‘‘writer’’ < schrijf ‘‘to write’’

Instrument maai-er ‘‘mower’’ < maai ‘‘to mow’’

Event miss-er ‘‘failure’’ < miss ‘‘to miss’’

Patient

bijsluit-er ‘‘instruction’’ < bijsluit ‘‘to enclose (with

medication)’’

The variation in interpretation for diVerent words of the same morpho-

logical type is also quite spectacular in the case of diminutives. Starting

from the basic meaning small, one may get to quite a number of interpret-

ational types. Consider the following examples of denominal diminutive

nouns in Dutch that serve to illustrate the wide variety of meaning

contributions associated with the diminutive suYx. Note that some of

these diminutive nouns are themselves polysemous:

222 interfaces

(30) kind ‘‘child’’ kind-je ‘‘small child, darling’’

broek ‘‘trousers’’ broek-je ‘‘pants’’

Geert ‘‘boy’s name’’ Geert-je ‘‘girl’s name’’

bier ‘‘beer’’ bier-tje ‘‘glass of beer’’

man ‘‘man’’ mann-e-tje ‘‘small man, little boy, unoYcial worker’’

auto ‘‘car’’ autoo-tje ‘‘small, inexpensive car, toy car’’

Cross-linguistically, the basic meanings of diminutives are those of ‘‘child’’

or ‘‘small’’, two closely related meanings that relate to size. The meaning

component ‘‘child’’ extends into notions such as aVection, sympathy, and

pet, that is, there is often a meaning component of endearment involved.

The notion ‘‘small’’ may also be associated with contempt, as when one

refers to a small and cheap car as an autootje. This second cluster of

meanings might be subsumed under the notion ‘attitude’ or ‘evaluation’.

Thus, diminutive morphology belongs to the semantic category of evalu-

ative morphology. The use of diminutive morphology for the creation of

female names probably reXects the idea that women are physically smaller

than men. The diVerent meanings of the diminutive can be related by

conceiving of its semantics as a radial category. A radial category has one

or more core meanings from which the other meanings derive through sem-

antic extension mechanisms such as metaphor. The structure in Figure 9.1

is a reduced version of what Jurafsky (1996: 542) considers as the universal

structure of the semantic category ‘diminutive’. The Dutch examples given

above are instantiations of the diVerent semantic categories in this scheme.

In Figure 9.1 related core meanings are given in bold. The diminutive noun

biertje in (30) is an example of the partitive use of the diminutive, it denotes

a part of a container with beer. The extensions on the bottom line of this

scheme are typically pragmatic in nature: since small children tend to

evoke aVection, these feelings can be transferred to other categories and

situations.

Fig. 9.1 The radial semantic structure of diminutives


In many cases the use of a diminutive suYx has a strong pragmatic

rationale, instead of just qualifying something as small. Therefore, Dressler

and Barberesi (1994) interpret the evaluative nature of the diminutive by

considering the pragmatic feature [non-serious] as its basic meaning. It is

often used in ‘motherese’, the language addressed to small children, as in

the following example from Viennese German (Dressler and Barberesi

1994: 106):

(31) Gut-i gut-i bist doch g’scheit-i g’scheit-i

good-dim good-dim are particle clever-dim clever-dim

‘‘Good-y good-y, you are a nice little clever child’’

The following observation by Wierzbicka (1991: 55), cited by Dressler and

Barbaresi (1994), on the use of the diminutive suYx in English is also very

relevant for grasping the role of diminutives:

(32) ‘Thus calling mosquitoes mozzies, the speaker is good-humouredly dismiss-

ing the problem: he thinks of mozzies as small (but not endearing), and

expects that the addressee would share this attitude’.

This is the kind of euphemistic use of the diminutive that also occurs when

a Dutch speaker asks: Hebt U nog een minuutje voor me ‘‘Do you have a

minute-dim for me?’’. You can be pretty sure that the conversation will take

more than a minute.

Another pragmatic role formorphology is the use of honoriWcs that express

diVerent degrees of politeness, and may also reXect the position in the social

hierarchy of speaker and addressee. In Japanese, for instance, the predicate

of a subject receives a special honoriWc form if the speaker addresses a

socially superior person. A relatively simple example of the politeness use of

the honoriWc suYx -masi in Japanese is given in (33) (Harada 1976: 502):

(33) a. Ame ga hut-ta

rain subj fall-past

‘‘It rained’’

b. Ame ga huri-masi-ta

rain subj fall-honor-past

‘‘It rained’’ (polite speech)

These examples show that the interpretation and use of complex words is

not only a matter of cognitive principles and mechanisms, but also has a

pragmatic dimension.

224 interfaces

A particular interpretation of a complex word may be the established

one, as is often the case. This is referred to as lexicalization. It does not

mean necessarily that such a word has become irregular, or that other

interpretations are now impossible (recall the example of diner), but simply

that a particular interpretation of a word has become part of the lexical

norm of the language community involved. Advertisements make use of

this. For instance, in an advertisement for convertible cars the Dutch

adjective dak-loos ‘‘lit. roof-less, homeless’’ was used in the sentence Wij

willen U graag dakloos maken ‘‘We would like to make you homeless/

rooXess’’, in which the literal interpretation ‘‘roof-less’’ is reactivated,

thus playing with the contrast between poor, homeless people, and people

who can aVord to buy a convertible type of car.

Summary

The semantic interpretation of complex words is governed by the general

principle of compositionality. Correspondence rules specify relationships

between formal structure and semantic interpretation of complex words.

The semantic scope of morpho-syntactic properties may be larger than the

word on which they are marked, as is the case for tense and mood proper-

ties. The semantic interpretation of a word may be further enriched by

conceptualization rules.

Pragmatic principles, knowledge of the world and of the context in which

a word is used, must also be invoked to arrive at a proper interpretation

and use of complex words.

The semantic interpretation of complex words and the semantic eVects of

morphological operations may have implications for the syntactic valency

of complex words, since the syntactic valency of a word reXects its semantic

properties.

Polysemy of aYxes and of individual complex words is a pervasive

phenomenon in the realm of complex words. It is the eVect of semantic

extension mechanisms such as metaphor and metonymy, and can be under-

stood more generally in terms of domain shift chains.

The meaning contribution of a particular aYx may be strongly prag-

matic in nature, as is the case for diminutives and honoriWcs.


Questions

1. Specify the semantic relationship between the two constituents of thefollowing English compounds: horse doctor, tree doctor, voodoo doctor,hospital doctor.

2. Identify for each of the following complex words which semantic extensionmechanism is involved in their interpretation: beehive, bottleneck, conven-tion, cliffhanger, holder, opposition, printer.

3. Which meaning component of the head noun is modified by the precedingadjective in the following NPs: a rural policeman, a Martian expedition, anearly riser, an individual decision?

4. Consider the following cases of adjective reduplication inMalagasy (Keenanand Polinsky 1998: 571):

fotsy ‘‘white’’ fotsyfotsy ‘‘whitish’’maimbo ‘‘stinky’’ maimbomaimbo ‘‘somewhat stinky’’hafa ‘‘different’’ hafahafa ‘‘somewhat different’’lo ‘‘rotten’’ lolo ‘‘somewhat rotten’’

What conclusion can be drawn from these data as to the universality ofcorrespondence rule (8)?

5. In the following sentence with a deverbal noun in -ion the use of a prepos-itional by-phrase leads to ungrammaticality:

John was the selection of/*by the committee

Why does the use of by lead to ungrammaticality?

6. Consider the following Turkish noun phrase (Kornfilt 1997: 225):

Turk-ler-in Istanbul-u feth-iTurk-PL-GEN Istanbul-ACC conquest-3SG‘‘The Turks’ conquest of Istanbul’’

How might it be explained that the noun Istanbul receives accusative casehere from a (deverbal) noun?

7. Dutch has a compound parkeergarage ‘‘lit. parking garage, multi-storey/underground car park’’. To what extent is this compound a violation of theNon-Redundancy Constraint, given the following range of meanings of theDutch word garage: ‘‘garage, service-station’’?

8. Specify the semantic scope of the suffix -ible in the following sentence:

This number is divisible by three.

9. The following English deverbal nouns in -ing are all polysemous. Determinefor each of them how their different meanings relate to each other: blessing,crossing, drawing, dwelling, opening.

226 interfaces

10. Some languages allow for double diminutives in which the root ofthe word is followed by two diminutive suffixes, as in Polish kot-ec-ek‘‘cat-DIM-DIM, dear pussy’’, and Afrikaans huis-ie-tjie ‘‘house-DIM-DIM, dearlittle house’’.

Explain why it is not so odd from the semantic point of view to use twodiminutive suffixes in one word.

Further reading

Conceptualization rules are discussed in Jackendoff (1983, 2002). The inter-pretation of English conversion verbs is discussed in Clark and Clark (1979) andAronoff (1980).The semantic resolution of the bracketing paradox issue is dealt with in more

detail in Spencer (1988) and Beard (1991).A survey of phenomena of inheritance and nominalizations, and discussions

of their proper analysis is found in Comrie and Thompson (1985), Hoekstra(1986), Grimshaw (1990), and Koptjevskaja-Tamm (1993). The analysis ofverbal compounds is discussed in Selkirk (1982), Botha (1984), Hoeksema(1985), and Booij (1988).The Non-Redundancy Constraint is discussed in Ackerman and Goldberg

(1996), and Goldberg and Ackerman (2001).Cruse (2000) deals with polysemy in general. The polysemy of English and

Dutch -er-nouns is discussed in Booij (1986) and in Booij and Lieber (2004).



Part VMorphology and Mind


10Morphology and

psycholinguistics

10.1 Morphology and mind

The previous chapters provided an introduction to morphological phe-

nomena, and discussed how these phenomena should be accounted for in

the grammar of natural languages. In this chapter, we will see how mor-

phological knowledge (knowledge of complex words and of morphological

rules) is represented in the human mind and how it is used in language

processing. This is an important topic because the mental representation of

morphological knowledge is a battle ground for diVerent theories about the

nature of linguistic rules, as we will see below. A related topic is the balance

between storage and computation. Knowledge in a particular domain of

human cognition always comprises both storage of information and the

ability to compute new information. For instance, when we want to use the

word books, we have two ways to do this: either we retrieve this plural form

of book from our lexical memory, or we create it on line, by adding the

plural suYx -s to the stem book. What determines the choice between these

two routes? Empirical research of such questions may thus help us to get

10.1 Morphology and mind 231

10.2 The mental lexicon 232

10.3 Acquisition of morphology 236

10.4 Sources of evidence 240

10.5 Models of morphological

knowledge 242

10.6 Morphological

processing 249

Summary 251

Questions 251

Further reading 254

a better understanding of the nature of human cognitive capacities.

Morphology may therefore be qualiWed as a window on the human mind.

In this chapter the empirical domains of psycholinguistic research will be

reviewed: the mental lexicon (‘the dictionary in your head’), the acquisition

of morphological knowledge, and the role of this knowledge in language

perception and production. In the next chapter, on morphological change,

we will see that language change also provides information about the

mental representation of morphological knowledge.

10.2 The mental lexicon

It is quite clear that the lexical knowledge that one has of one’s native

language does not have the format of a dictionary. First, the number of

lexical entries in a good dictionary is much higher than that in our individ-

ual mental list of words. There are many words that most speakers do not

know. You can compute the size of your own vocabulary by checking, on

the basis of a representative sample of words from a good dictionary of

your native language, how many of the words in that sample you know.

Such a sample may be constructed in a number of ways, for example by

taking the Wrst word of each page of that dictionary. This will give you a

percentage p of known words. If the dictionary contains nwords, the size of

your mental lexicon is p times n. Adult speakers of English with a higher

education might know up to 50,000 word types, and sometimes more, but

it is certainly a smaller set than the whole English vocabulary, which

comprises hundreds of thousands of words. This estimate concerns the

passive vocabulary, the number of words that you understand. The active

vocabulary, the set of words one uses in language production is much

smaller.

A dictionary is conservative by nature, and hence it will contain words

from the past that nobody uses any more. Each new edition of a printed

dictionary will contain new entries, but will also have deleted a number of

words from the previous edition that have become obsolete. Notwith-

standing this kind of updating, your mental lexicon will always be ahead

of the dictionary, and contains a substantial number of words that are not

listed in dictionaries. New words (neologisms) are coined continuously, and

dictionaries always lag behind. Moreover, the editors of dictionaries use a

threshold for the listing of words: a new word must have a certain degree of

232 morphology and mind

permanence in the language use of more than one speaker before it gets an

entry in a dictionary. Language users do not have such a threshold for their

mental lexicon.

A second diVerence between a dictionary and the mental lexicon is that

words in the mental lexicon bear a number of relationships to each other.

Words with similar meanings or similar phonological forms appear to be

related in the mental lexicon, as can be concluded from speech errors. For

instance, someone who wanted to talk about the sympathy of a musical

conductor used the word symphony instead of sympathy. The similarity in

form between these two words, and the semantic relatedness of symphony

and conductor must have played a role in this speech error. In a dictionary,

on the other hand, semantic relations between words are usually not

expressed directly. Thus, we may conceive of the mental lexicon as a

multidimensional web of words, with all kinds of connections between

those words: semantic similarities, phonological similarities, and morpho-

logical relationships. In a dictionary, on the other hand, words have one

type of relationship only, that of alphabetical ordering. This ordering is

based on the degree of orthographical similarity between words.

A third diVerence between a dictionary and a mental lexicon is that the

latter also stores information about the frequency with which you come

across a word. Linguists may compute the frequency of words on the basis

of large corpora of actual language use. Frequency counts indicate how

many tokens are found for each word type in a particular corpus. For

instance, function words such as determiners have a relatively high

frequency, and the probability that the next word that you come across

when reading an English text is the word the is much higher than that it is

the word ubiquitous. Within the set of English adjectives, the word nice

is used far more frequently than the word opaque.

At Wrst sight, this kind of frequency information may be seen as external

to the human mind and without relevance for the mental representations of

words. Language users do not have a speciWc frequency number for

each word in their mental lexicon, as if you are continuously counting

tokens of words when using language. However, speakers are able to

estimate the relative frequencies of words based on their continuous

exposure to language use. When you ask an English speaker if nice is less

or more frequent than ubiquitous, he/she is certainly able to come up with

the right answer. Moreover, a word with a relatively high frequency of

occurrence is more easily recognized and retrieved than a word with a

morphology and psycholinguistics 233

relatively low frequency. A possible psycholinguistic interpretation of this

latter frequency eVect is to assume that frequency of exposure to a word

correlates with the activation level of that word in the mental lexicon. A

word with a higher activation level is activated faster in language processing.

The standard way of Wnding out about such frequency eVects is that of

psychological experimentation, for instance by means of a lexical decision

task. The participants in such an experiment may be asked to determine

if a sequence of letters shown on a screen is a word of their language. There

are two buttons, one for yes and one for no, and the time interval between

the letter sequence appearing on the screen and one of the buttons being

pushed, the response latency, is measured in milliseconds. The prediction

then is that the higher the frequency of a word as determined by the

experimenter on the basis of a corpus, the shorter the time it takes to

come to a decision. This eVect has been found time and again, and it is

very robust indeed. It is referred to as the frequency eVect.

A related eVect is the cumulative frequency eVect. If, for instance, the

singular form of a noun is highly frequent, this will also have a frequency

eVect on its plural form, even when that plural form itself is not particularly

frequent: the token frequencies of related inXectional word forms contrib-

ute to the activation level of each of them. This is called the base frequency

eVect. Another example is the cumulative root frequency eVect: the summed

frequencies of all words with the same root also appear to aVect response

latencies of individual complex words containing that root.

An example of the calculation of these frequencies for the English verb to

calculate is given in Table 10.1. These counts are based on the CELEX

Table 10.1. Frequency data for calculate

InXected forms Token frequency Family members Base frequency

calculate 108 calculate 574

calculated 340 calculable 4

calculates 21 calculation 343

calculating 105 calculator 89

calculus 50

incalculable 26

incalculably 1

miscalculate 5

miscalculation 25

Source: De Jong et al. 2000: 329.


database, which contains token frequency counts for Dutch, English, and

German words (Baayen et al. 1995). The base frequency of calculate is

the summed token frequencies of its four inXectional forms, 574. The

cumulative root frequency of calculate is the sum of the frequencies of all

the words in its family, 1,117.

The standard psychological interpretation of the base frequency eVect

and the cumulative frequency eVect is that of activation spreading. Activa-

tion can be spread from one word or word form to a related one in the

mental lexicon, thus raising the activation levels of related words. There-

fore, the response latencies of words of the same family in lexical decision

tasks will be aVected by their forming part of that family.

Recently, it has been discovered that response latencies for simplex words

are also aVected by the family size of these words, that is, the number of

morphologically related word types. The larger the family size, the faster

lexical decisions on simplex words will be made. In the case of calculate, the

family size is eight, since there are eight morphologically related word types.

This eVect on response latencies is called the family size eVect (Schreuder

and Baayen 1997).

It takes a number of years to build up your mental lexicon, and during

your whole life you will continue adding and losing words. It does not

suYce just to memorize the simplex words of your native language, and

either decompose (in case of perception) or construct (in case of produc-

tion) complex words on the spot. First, complex words have to be memor-

ized if they have an unpredictable property, for instance an idiosyncratic

meaning aspect. Second, even though a complex word is completely

regular, it has to be memorized if it is the conventional name for a particu-

lar thing. The device we call a typewriter in English could also have

been called a writing machine. Both words have transparent semantic

interpretations, and are formed according to the English rule of nominal

compounding. We have to remember that it is the Wrst word that is the

conventional name for this device, and therefore it must be stored in lexical

memory. Moreover, we have to memorize that it cannot be used as a

synonym for typist.

There is also a psychological reason why we store regular complex

words. There are two ways in which linguistic information can be accessed,

either through computation or through storage. If we come across a

complex word that we have to recognize and interpret correctly (the task

of word recognition) there are two ways of reaching our goal. If we use


computation, we Wrst decompose the complex word into its constituent

morphemes, and then retrieve the meaning of these morphemes from the

entries in our mental lexicon. In the storage scenario, the complex word as

such is stored in our mental lexicon, and word recognition is performed by

matching the perceived word with its corresponding entry in our mental

lexicon. The Wrst scenario is the only possible one for complex words that

we never came across before. For words that we already know, there are

two routes: retrieval from memory or computation. Our human memory

has such a vast capacity that there is no reason to assume that our

lexical memory is redundancy-free. That is, we have the capacity to store

information that might also be computed. The advantage of using the

retrieval-from-memory route is its speed: it will take less processing time

than the computing route, in particular when the stored word has a high

frequency of occurrence, and thus a high level of activation. Hence, an

important part of psycholinguistic research is Wnding out about the proper

balance between storage and computation in morphology.

A clear proof of this point is that some of the regular plural forms of

Dutch and Italian nouns have been found to be stored. InXection is

typically associated with rules. Dictionaries quite often do not give the

regular plural forms of nouns because they can be formed by the language

user who knows the relevant inXectional rules. Yet, there are frequency

eVects for plural nouns: plural nouns with a high frequency of occurrence

are recognized faster in lexical decision tasks than plural nouns with a

low frequency of occurrence, even if the words do not diVer in their

base frequency. This diVerence in activation level shows that these high

frequency regular forms must be stored in the lexicon (Baayen, Burani, and

Schreuder 1997; Baayen, Dijkstra, and Schreuder 1997). Yet, speakers also

know the rule for plural nouns, since they can make plural forms that they

never came across before.

10.3 Acquisition of morphology

How does a child acquire the morphological system, the set of morpho-

logical rules of his or her mother tongue? Morphological rules have to be

discovered on the basis of words that are formed according to these rules.

Therefore, the Wrst stage in the acquisition of morphology is the acquisition

of individual complex words. Based on this knowledge, the child will be


able to use morphologically complex words correctly without making use

of morphological knowledge, by retrieving them from memory. Next, (s)he

may discover certain recurring properties of, for instance, plural nouns in

English, and conclude that they are formed by adding the ending -s (with

the allomorphs [s], [z], and [1z]) to the stem. So the child is able to apply the

rule for English plural nouns to new cases, and create plural nouns that

(s)he has never heard before.

In cases where rules have exceptions, a further reWnement is in order,

and three stages have to be distinguished. These three stages of morpho-

logical acquisition have been argued for in a classic paper by Berko

(1958). In the Wrst stage, children learn a number of, for instance, past

tense forms of English verbs by rote. Hence their production of past tense

forms of both regular and irregular verbs (asked, went) is correct. In the

second stage, children acquiring English have discovered the rule for past

tense forms, but will also apply the rule to the class of irregular verbs.

Hence, they will produce the correct form asked, but the incorrect form

goed instead of the correct went, a case of overgeneralization. In the third

stage, both the rule and the set of exceptions have been acquired, and

thus children will produce asked and went. So the learning process has

the shape of a U-curve: the number of correct past tense verbs in stage I

decreases in stage II, and increases again in stage III. This learning curve

has been found for many languages. In Romance languages, with a

number of diVerent conjugations for verbs, it is the default conjugation

that is overgeneralized. In French, for instance, the conjugation of verbs

in -er is the conjugation with the highest type frequency and the default

one. It is that conjugation that verbs might incorrectly be assigned to in

stage II.

Overgeneralization in certain stages of language acquisition nicely

reveals that discovering the regularities is part and parcel of language

acquisition. Another example is the way in which Dutch children treat

the plural of loan words from English such as Xat and tram. Adult speakers

of Northern Dutch use the English plural forms Xats and trams. These are

irregular forms since normally the Dutch plural suYx -s is not used after

monosyllabic consonant-Wnal nouns. Children, however, may use the

plural forms Xatten and trammen. These are the regular forms, since

Dutch monosyllabic nouns that end in a consonant take -en as their plural

ending. Children may use these regular forms because their mental lexicon

does not yet contain the irregular plural forms Xats and trams.


Onemay wonder why the stored form went does not block the creation of

goed in the second stage, where overgeneralization applies. After all, this

form is kept in memory at an early age due to its high frequency, and it may

therefore be expected to block the creation of goed. It is probably the case

that children in that stage of acquisition consider went as a verb of its own,

and not as a (suppletive) form of to go. This is supported by the fact that

some children create the past tense form wented.

The persistence of irregular or non-default word forms in a language has

to do with their high frequency of use. Many of the stem-alternating or

irregular verbs of the Germanic languages have a high frequency of use.

Hence, these forms are entrenched very strongly in lexical memory, and will

therefore not be replaced with the competing regular suYxed past tense

forms. Frequency also plays a role in the preservation of suppletion. Words

that have suppletive forms, such as the adjective bad with the suppletive

form worse, are often words with a high frequency of use.

In languages with case marking, children will use case-marked word

forms from very early on because it is words in their surface forms (not

stems) that form the basis of acquisition. For nouns, the nominative form

is usually the default case form, used in all syntactic contexts. Children can

discover the role of case marking in syntactic contexts at a very early age.

For example, Russian children already use accusative nouns after transitive

verbs before the age of 2 (Clark and Berman 2004).

In the domain of word-formation children discover the building

principles for complex words quite soon. We know this because they are

able to coin new words themselves at a very early age. Children coin words

for concepts for which there are already established words. For adults,

established words have priority above new coinings, in accordance with the

principle of conventionality that gives priority to established words, and

thus blocks the creation of newwords that are synonymous to existing ones.

Children are less hampered in their word-formation creativity by this

principle because their mental lexicon does not yet contain that many

words, and still has many lexical gaps. For instance, my daughter Suzanne

created the Dutch compound oorlogsman ‘‘war man’’ with the meaning

‘‘soldier’’ instead of using the established word soldaat ‘‘soldier’’. An

English example is the use of the compound sky car instead of the

established compound aeroplane. This behaviour is thus parallel to the case

of overgeneralization with the past tense form goed. These observations

show that children do not just memorize the complex words they are


exposed to, but are able to discover the structural principles behind those

complex words. The development of this ability in children is also mani-

fested in the creation of new compounds for complex concepts without

corresponding established words in their native language, such as lion book

‘‘book about lions’’, baby-bottle ‘‘bottle the child had used as a baby’’, and

hole-sack ‘‘sack with holes in it’’ (Clark 1993: 117).

Compounding is acquired relatively early because it complies with two

principles, those of transparency and simplicity. The meaning of com-

pounds can be related very easily to those of its constituent words, and

hence their meaning is transparent. Moreover, the form of the constituent

elements is not really changed when they are part of compounds, and thus

conforms to the requirement of simplicity (Clark 1993). This stands in

contrast to word pairs like magic–magician where the base word magic,

with initial stress, has a diVerent phonological shape when it is part of

magician: stress on the second syllable instead of the Wrst, and a diVerent

root-Wnal consonant. It is therefore harder to discover the morphological

structure of magician. The principles of transparency and simplicity are

also obeyed in conversion. Consider the following coinages by children

of 2–3 years old: (Clark 1993: 117):

(1) to Xag ‘‘to wave with a Xag’’

to Wre ‘‘to light (a candle)’’

to bell ‘‘to ring’’

to gun ‘‘to shoot’’

to dust ‘‘to get dust on’’

The meaning of these verbs can be easily determined on the basis of the

meaning of the base noun.Theydenote activities inwhich the base nounplays

a role, and which can be further identiWed on the basis of the context in which

these verbs are used. Hence, their meanings are transparent. In conversion,

the forms of the words remain the same. Thus, noun–verb conversion com-

plies with the requirements of transparency and simplicity.

A third factor involved in the acquisition of morphology by children is

that of productivity. In English both compounding and conversion are very

productive. This is reXected by the fact that children acquire and use these

types of word-formation at an early age. In French, on the other hand,

compounding and conversion are not productive, and derivation is used

instead. Thus children acquiring French will use suYxation earlier and on

a larger scale than children acquiring English. In Germanic languages,


productive aYxes such as deverbal -er for the coinage of agent and instru-

ment nouns (in English, German, and Dutch) are acquired earlier than less

productive ones.

Twomorphological processes in the same language may be both transpar-

ent, andyet diVer inproductivity in adults.That is, transparency is a necessary

condition for productivity, but not a suYcient one. For instance, the English

suYxes -ness and -th diVer in that new de-adjectival nouns in -th are almost

never created.Yet, both typesofde-adjectival nouns conform to theprinciples

of transparency and simplicity. Clearly, children have to discover these

diVerences in productivity. They have to know themorphological preferences

of the language community when they are going to make new words.

Since children discover morphological processes on the basis of the words

they are exposed to, they must be able to compare and analyse words, and

assign them morphological structure if possible. This is clear from word

forms like goed, as discussed above. Moreover, children appear to analyse

words consciously, and ask questions like: Does cornXakes have corn in it?

(Clark 1993: 40). This analytic activity also manifests itself in what we

may call folk etymology: children may come up with a motivation for a

particular word that is incorrect. An example from Dutch is that a child

interpreted the name of the province of Friesland as meaning ‘‘land where it

is always freezing’’ (Dutch has a verb vries ‘‘to freeze’’ that has nothing to do

with the Wrst constituent of the compound Friesland).

10.4 Sources of evidence

How can we determine how morphological information is represented

in themind? There are twomain sources of evidence: experimental and natur-

alistic data. Psychologists use experiments, and in the realm of morphology

lexical decision tasks are the most widely used way of probing into the

mental representation of morphology. Lexical decision tasks are often used

in combination with priming, the prior presentation of another word. This

means that the subject who has to make a lexical decision Wrst receives some

other information.

An example of priming is the following: if one has to decide if calculation

is a word of English, the response latency will be reduced if, before this task

is performed, the same word calculation is presented to the eye or the ear of

the subject, a case of identity priming. This eVect suggests that the word


calculation receives a higher level of activation by previous access to the

same word. Other ways of priming are phonological priming (the prime

word is phonologically similar to the target word), and semantic priming

(with a semantically similar word as the prime). In the case of the target

word calculation, the words calcium and computation may function as

phonological and semantic primes. If these primes reduce response laten-

cies for the target word, one can conclude that words in the mental lexicon

are connected to words that are either phonologically or semantically

similar. In morphological priming, a word that is morphologically related

to the target word is used as a prime. For instance, calculatewill function as

a prime for calculation. As this example shows, it may be that the latter

kind of priming is in fact a combination of phonological and semantic

priming, and not an independent phenomenon.

Naturalistic data are the second source of evidence for theories of

morphology in the mind. These are data concerning the actual behaviour

of language users. When we observe that children coin new words or word

forms that they have not come across before, this forms naturalistic data

that may be interpreted as evidence for the children’s having acquired

morphological rules. The same applies to the language use of adults. We

can conclude that a morphological pattern has the status of productive rule if

new instantiations of that pattern are found in the actual language use of

native speakers.

Speech errors (which can also be elicited in experiments) are another

instance of naturalistic data. Consider the following slip of the tongue

(Aitchison 2004):

(2) Take the freezes out of the steaker (take the steaks out of the freezer)

In this sentence, the roots freeze and steak have been exchanged erroneously.

This speech error shows, Wrst of all, that the correct allomorph of the English

plural ending canbe computedby rule: theplural suYx for steak is [s],whereas

that for freeze is [z], because the stem ends in /z. Second, this exchange

of roots bears on the issue of the lexical representation of morphological

structure. The word freezer is a complex word that is presumably stored as

such in the mental lexicon, given its speciWc conventional meaning (a storage

device in which things are kept frozen in order to keep them), and its

frequency.Yet, as this speecherror shows, its internalmorphological structure

is still accessible. Therefore, we conclude that complex words can be stored in

lexical memory with their morphological structure still present.


A similar example from Dutch is the following speech error:

(3) Met excuses voor het verbind-en van de verbrek-ing (het verbrek-en van de

verbind-ing)

‘‘With apologies for the connect-INF of the interrupt-ion’’

Crucially, in both (2) and (3) it is morphological units that are exchanged,

and not phonological units such as syllables. The word freezer, for instance,

consists of the syllables (free)s and (zer)s, but it is the sequence of sounds

freez, a morphological unit, that is exchanged. In example (3) the morpho-

logical units -en and -ing are exchanged. This exchange cannot be a matter

of syllables since the syllable division of the target words is as follows:

ver.bre.ken, ver.bin.ding. If we exchanged syllables, we would get the words

verbinken and verbreding.

A last source of naturalistic data to be mentioned here is that of language

pathology. The language behaviour of people with a language deWcit, such

as aphasia, may throw light on the mental representation of morphology.

Aphasia is a language impairment resulting from damage to the brain, in

most cases due to a cardiovascular accident. Aphasics that suVer from

agrammatism (the inability to use rules), usually referred to as Broca

aphasics, may not be able to produce correct plural forms of nouns, except

for nouns with a high frequency plural form such as the word eyes (Aitch-

ison 2004). This is evidence for the position that high-frequency plural

nouns are retrieved from lexical memory, whereas low-frequency plural

forms are computed on the spot, and may not be stored. In an investigation

of three German aphasic patients with agrammatism it was found that their

syntactic abilities were severely impaired, whereas their ability to use inXec-

tional morphology was still intact (de Bleser and Bayer 1988). This fact can

be interpreted as evidence for the position that syntactic rules belong to a

diVerent module of the grammar than the rules of inXectional morphology.

10.5 Models of morphological knowledge

The observed ability of native speakers to coin new words or word forms

is the basic argument for assuming that morphological knowledge encom-

passes more than storage of the complex words that language users are

exposed to. There must be mechanisms in the mind that enable us to extend

the set of complex words in a language. For languages with a rich system


of inXectional morphology such as Turkish, where each word may have

hundreds or thousands of forms, it would even be quite absurd to assume

that all these forms are memorized as such. The memory load for such

languages can be reduced considerably by making use of rules. A possible

model is that language users acquire morphological rules by constructing

abstract rules or templates on the basis of their lexical knowledge. How-

ever, there are also morphological patterns that are not productive, but can

still be called patterns. The stock example is the set of English past tense

verb forms. Regular past tense forms are created by suYxing -ed to the

stem. On the other hand there are a number of stem-changing verbs where

the past tense form and the past participle are marked by having a diVerent

vowel than the present tense form. For instance, we have an -ing/k -ang/k

-ung/k pattern for verbs such as the following:

(4) ring, sing, spring, drink, shrink, sink, stink

The class of verbs with this kind of vowel alternation will normally not be

extended. Yet, it is clear that native speakers of English are able to recog-

nize the abstract pattern involved. It has been observed that some speakers

tend to also inXect the verb to bring according to this pattern, resulting in

the forms brang and brung. If people are asked to make the past tense forms

of non-existing verbs with a similar phonological make-up as the words in

(4), such as spling, they tend to come up with either splang or splung.

Moreover, verbs that were once regular have changed into irregular ones,

such as the English verb to dive (with the past tense form dove instead of

dived in American English).

These facts have led a number of linguists to defend a dual system model

of morphological knowledge (cf. Pinker 1999). The past tense and partici-

pial forms of regular English verbs are not stored in lexical memory,

but always created by rule. The irregular forms, on the other hand, are

stored in memory. These stored irregular forms are linked to each other in

an associative way, and thus the language user will be able to discover

similarity patterns such as -ing/k -ang/k -ung/k for the set of verbs in (4).

This explains why the pattern may be extended incidentally to similar verbs

such as to bring.

The dual system theory has also been defended based on a type of

evidence that has not been mentioned yet: neurological evidence. Jaeger

et al. (1996) have argued that regular and irregular verbs in English are

processed by diVerent neural systems. The processing location can be


determined by PET-scans. PET stands for Positron Emission Tomography.

The assumption underlying this technique is that changes in brain activity

during the performance of a task are associated with changes in blood Xow

in the diVerent regions of the brain. Blood Xow can be made visible by

injecting a participant’s veins with a radioactive Xuid. Thus, the blood Xow

can be traced through diVerences in radioactivity in the diVerent parts of

the brain: the more blood Xow, the more radioactivity. This radioactivity is

caused by the emission of positrons. Participants had to read or speak past

tense forms (regular and irregular ones). Jaeger et al. (1996) found that

regular forms are processed faster than irregular forms, and also that the

irregular past task activated signiWcantly larger areas of the brain than

the regular past task.

The dual system theory in the form presented above thus correlates the

distinction between regular and irregular morphology with the distinction

between computation and storage: regular forms are computed on the spot,

whereas irregular forms are stored. This theory predicts that frequency

eVects (diVerences in frequency correlating with diVerences in processing

speed) will only be found for irregular forms, and not for regular forms,

although the latter have a high frequency of use. Some linguists have

claimed this prediction to be correct (for instance, Clahsen 1999 for

German verb forms).

A problem for this theory is that in a number of studies frequency eVects

have been found for regular inXectional forms as well. These Wndings do not

necessarily contradict the idea of an opposition between rules versus

associative patterns, but imply that a correct model has to allow for the

storage of regular inXectional forms of relatively high frequency besides

storage of irregular forms. Remember that there is no logical contradiction

between a process being regular and productive, and the storage of some of

its outputs: our mental lexicon is not redundancy-free. Even if one accepts

the storage of regular forms, the dual route theory might be maintained.

The crucial property of regular forms is not that they cannot be stored, but

that they can be computed if necessary, unlike the irregular forms.

The dual system theory raises the question of how to distinguish between

rules and associative patterns. The crucial diVerence between these two

notions is that rules make use of variables, whereas associative patterns

refer to speciWc properties of the words involved, for instance their speciWc

phonological make-up. That is, unlike associative patterns, rules also apply

to types of input that the language user has not been exposed to.


This reasoning can be illustrated by a study of Modern Hebrew by Berent

et al. (2002). As we saw in Chapter 2, Hebrew makes use of abstract

consonantal patterns to deWne the triconsonantal roots that form the

basis of Hebrew morphology. A general constraint on these triconsonantal

roots is that the Wrst two cannot be identical, but the last two can. For

instance, roots with the pattern smm are well formed, unlike roots of the

form ssm. If you do not believe in abstract rules, you may claim that this

generalization is of a statistical nature, a strong tendency that native

speakers may discover on the basis of frequent exposure to words with

such roots. Speakers of Hebrew were asked to evaluate the acceptability of

non-existing roots in which either the Wrst two or the last two are identical.

The crucial Wnding of Berent et al. (2002) is that native speakers only accept

roots of the second type (the last two consonants identical), even with

foreign consonants such as the English phoneme /y/, but systematically

reject roots where the Wrst two consonants are identical. In other words,

speakers are able to make systematic judgements although they have never

been exposed before to consonantal roots with this foreign phoneme.

Berent et al. (2002) therefore argue that this constraint on identical con-

sonants is not a statistical generalization about the set of existing roots, but

a real abstract constraint, a constraint with a variable for ‘‘consonant’’.

The dual system approach to morphological knowledge stands in op-

position to a class of theories with the common denominator that they are

single system theories. These theories might be called rule-less theories

because they do not make a rigid distinction between rules on the one

hand, and (lexical) representations on the other. The most inXuential single

system theory is the connectionist approach, applied to the domain of

English past tense forms in a famous study by Rumelhart and McClelland

(1986). This approach uses the model of the neural network to represent

English speakers’ knowledge of past tense forms. Neural networks model

how the brain encodes information. Information is encoded by means of

connections of varying strength between neurons. Such models may also be

used to model linguistic knowledge, although there is no logical necessity

that the architecture of knowledge systems mirrors the way in which

that knowledge is neurologically encoded directly. The basic idea of the

connectionist approach is that morphological knowledge is conceived of as

a pattern associator memory, a set of associations between pieces of

linguistic information. The strength of these associations varies with the

number of times that one has been exposed to the relevant information. In


the connectionist modelling of English past tense formation, the memory

contains present tense and past tense forms of words, or, more precisely,

their diVerent phonological components (these tense forms are not repre-

sented as units themselves). These components are associated to each other.

For instance, the phonological components of the input walk are associated

to that of the output walked. This set of associations has a morphological

label, for instance [past tense]. The set of associations between the phono-

logical features of the sounds in the sequence alk and those in the sequence

alked under the label [past tense] will also hold for the pair talk–talked.

Similarly, there will be a network of associations of considerable strength

between the phonological constituents of ing and ang that accounts for the

past tense forms of to sing and the other verbs listed in (4).

A small part of such a pattern of associations is represented in Figure

10.1. Each line is an association line between pieces of information. Each

input unit and each output units consists of a linear sequence of three pieces

of information. A piece of information is either a phonological feature or a

bracket. The left and right brackets [ ] indicate the beginning and the end of

a word. Each input unit thus speciWes a subsequence of three segments

(including the word boundaries) of words. For instance, the present tense

verb swim corresponds to the input unit ‘high nasal]’ since it ends in a

sequence of a [high] vowel and a [nasal] consonant followed by a right

bracket. Therefore, association line 1 applies to swim–swam. Indeed, swam

ends in a [back] vowel followed by a [nasal] consonant and a right bracket.

Association line 2 holds for the pair ban–banned. Association line 3 holds

for words with a word-initial cluster of two consonants, and hence relates

the word pairs swim–swam, sleep–slept, and bleep–bleeped. Association line

4 applies to sleep–slept, and also to bleep–bleeped: the stem forms of these

words end in a [high] vowel plus a stop consonant, and the past tense forms

Fig. 10.1 Some association patterns for English past tense forms (after

Pinker 1999: 105)


end in two stops. The second output unit, ‘nasal stop ]’ is also associated

with the Wrst input unit (association line 5) because there are pairs such as

trim–trimmed: the past tense form trimmed ends in a nasal stop sequence.

The association patterns and their strengths are developed by training

the model: it will be oVered a corpus of present and past tense forms.

The more frequent a particular association of pieces of information is,

the stronger that association will become. Once the model has been exposed

to a suYciently large number of data, it will also predict the past tense

forms of verbs that did not belong to its ‘‘training space’’. Most verbs

exhibit the pattern x�xþed for present and past tense. Yet, the high token

frequency of the past tense forms of irregular verbs results in great strength

of the associations between the phonological features sequences of ing/k

and ang/k in the domain [past tense] for the irregular English verbs. If the

input sing is fed into the network, the input units that correspond with parts

of the phonetic string of sing will be switched on. These input units then

activate the output units with which they have strong associations. This will

have the eVect that for the verb to sing the past tense form that is computed

by the model will be sang rather than singed. The verb trim, with diVerent

initial segments, will activate a diVerent set of strong input–output

associations, and thus produce the past tense form trimmed.

A second type of rule-less approach is the model proposed by Bybee

(1988, 1995, 2001). In her model, unlike what is claimed in the connection-

ist approach, individual inXectional forms of words are listed in the lexicon.

This also applies to past tense verb forms, both regular and irregular ones.

The regularities in the formal relationship between present tense and past

tense forms are captured by abstract schemas that can be constructed on

the basis of sets of words. For instance, the lexical representation of the

past tense forms played, spilled, spoiled, banned, and rammed will be linked

as in Figure 10.2. The correlation between the phonological and semantic

Fig. 10.2 Schema for English past tense forms (Bybee 2001: 23)


properties of these forms results in a morphological schema in which -ed

is connected to [past]. The basic point is that one may represent the relevant

generalization without necessarily extracting it as an abstract rule or

template independent from the individual words involved.

A third single system approach without rules is the analogy-based model.

To be sure, analogy has been recognized as a source of new complex forms

in all kinds of morphological models. In the case of analogy, a new word

is formed on the basis of a related existing word, without an abstract

morphological template being involved. For instance, the coinage of the

English word seascape is based on the word landscape (a borrowing from

Dutch), according to the following pattern of analogy:

(5) land : sea ¼ landscape : seascape

A similar example is the coinage of the word zweisam ‘‘being with some-

body else’’ in analogy to einsam ‘‘being alone’’ by the German philosopher

Martin Buber. In these cases it is an individual complex word that forms

the basis for the new word, and this kind of analogical word-formation has

an incidental character.

In recent theorizing, analogical models have been extended to larger sets

of data. These exemplar-based models presuppose that individual complex

words are stored in the lexicon. These words are exemplars of a particular

category, for instance of the noun–noun-compound category. There are no

abstract patterns or rules that deWne that category. New words are coined

in analogy to the set of stored words. An example of this kind of modelling

concerns the selection of a linking element in Dutch nominal compounds.

Many of these compounds have s or en as a linking element between the

two parts of the compounds; other compounds have no linking element at

all (represented here as ø). It is hard to come up with a rule that predicts

with absolute certainty which linking element should be chosen for a

new compound, and there is sometimes variation among native speakers.

Yet, there are tendencies that can be related to the distribution of linking

elements in existing Dutch compounds. The preferences of native speakers

for a particular linking element in a new compound can be predicted on the

basis of its degrees of similarity to existing compounds. This implies that

there is no rule that predicts the insertion of a linking element on the basis

of properties of the two constituents. Instead, a paradigmatic approach is

chosen in which the relation to other existing compounds plays a crucial

role, and preferences can be predicted.


Consider the following data concerning linking elements in Dutch

NN-compounds:

(6) schaap-s-kooi ‘‘sheepfold’’

schaap-s-vlees ‘‘mutton’’

schaap-ø-herder ‘‘shepherd’’

schap-en-poot ‘‘sheep’s leg’’

schap-en-vacht ‘‘sheep’s Xeece’’

schap-en-bout ‘‘leg of mutton’’

koei-en-oog ‘‘cow’s eye’’

paard-en-oog ‘‘horse’s eye’’

pauw-en-oog ‘‘peacock’s eye’’

varken-s-oog ‘‘pig’s eye’’

(the variation schaap/schap is only a matter of orthography). Given these

data, we might wonder which is the appropriate linking element for the

compound schaap-?-oog ‘‘sheep’s eye’’ (the question mark indicates the pos-

ition of the linking element). It is clear that the left constituent schaap allows

for all three linking elements, but en is themost frequent one.Moreover,most

constituents oog as heads of compounds are preceded by the linking element

en. Hence, the most probable linking element is en. It appears that speakers

tend to choose that linking element that is most common among the existing

compounds with that initial constituent. For instance, the choice of the

linking element s for a new compound leven? therapie ‘‘life therapy’’ appears

to be predictable on the basis of the existing compounds with leven as their

Wrst constituent, such as leven-s-vorm ‘‘life form’’ and leven-s-probleem ‘‘life

problem’’ (Krott et al. 2001). The second constituent may play a role as well,

as is illustrated here by oog that is usually though not always preceded by en.

This survey of diVerent models of morphological knowledge, and the

debate about these models illustrates that morphological phenomena play

an important role in the debate on the nature of linguistic knowledge.

Morphology is an important window on the mental representation of

natural language and the human mind.

10.6 Morphological processing

The insight that complex words are often stored as such in the lexicon raises

the question of how they are processed. There are two ways in which a

complex word can be processed, the direct route, and the indirect route.


Whenperceptionof complexwords is involved, thedirect routemeans thatwe

do not Wrst parse the complexword, but go directly to its representation in the

lexicon, in order to access its meaning. The indirect route means that a

complex word is parsed into constituent morphemes, and that its meaning is

computed after we have gained access to its constituent morphemes and their

meanings.

There are data that might be interpreted as showing that language users

try to parse words into morphological sub-constituents. For instance, when

speakers of Dutch are confronted with non-words in a lexical decision task,

it takes more time to make the correct decision when that word begins with

a syllable that could have been a preWx, compared to words where this does

not apply. Consider the following three non-words:

(7) aderibag, afgeblar, afbegepakt

The Wrst word does not contain a potential preWx of Dutch, the second

word begins with two preWxes (af-, ge-), and the third word begins with

three preWxes (af-, be-, ge-). Correlating with this diVerence, the last non-

word has been found to have the longest response latency, as is expected if

language users try to strip preWxes from words that are not in their lexicon.

The existence of two routes has led to the assumption of morphological

race models. In such models, both routes will be used for word recognition,

and compete with each other. If the word is not stored in lexical memory,

the indirect route is the only one that will lead to recognition, on the basis

of parsing. If the word is stored, the direct route will be faster than the

indirect route if the stored complex word has a high frequency, that is, a

high level of activation. For a complex word with a low frequency, on the

other hand, parsing is the obvious route. Both routes are followed, and one

of them will turn out to be the fastest, depending on the level of activation

of the complex word.

The parsing of complex words is also aVected by relative frequency: the

frequency of the complex word compared to that of its base word. If the

complex word has a lower frequency than the base word, this will make

parsing of that complex word easier and more eYcient than direct retrieval.

If the inverse situation obtains, direct access to the complex word will be

the most eYcient route (Hay 2001). For instance, the English word dazzle-

mentwill be parsed because its frequency is much lower than that of its base

dazzle, whereas government will be accessed as a whole because the

frequency of this word is higher than that of its verbal base govern. Note


that the meaning of government is not completely compositional, unlike

that of dazzlement. There is also a relation to productivity: the more

complex words with a certain aYx will be parsed, the more productive

the corresponding word-formation process will be, because parsing the

aYx will increase its activation level (Hay and Baayen 2002).

The morphological race model can also be used as a model of word

production. It helps us to understand why it is irregular forms of high

frequency that aremaintained in a language, whereas low-frequency irregular

forms disappearmore easily. Suppose we have tomake the past tense form of

to go.We can go directly to the lexicon, where the wordwentwill be retrieved.

Because went is a high-frequency form, this route will be faster than the

indirect route where the past tense form has to be created by rule, resulting

in the incorrect goed. The race model thus accounts for the blocking eVect of

high-frequency words on synonymous words that are computed by rule.

Summary

Children acquire the morphological system of their mother tongue on the

basis of exposure to complex words. At a relatively early age, they acquire

and use the competence to create new words themselves. Word-formation

processes that obey the constraints of transparency and simplicity are

acquired Wrst. Children also coin a high number of new words because

their creativity is not yet hampered by the conventional lexicon.

There are a number of ways in which we can Wnd out how morphological

knowledge is represented in the mind and used in language processing: by

studying naturalistic data (corpora of language use, speech errors, eVects of

language impairment, etc.) and experimental data (lexical decision tasks,

production tasks, etc.). Morphology is a battleWeld for competing models

of linguistic knowledge, and for discussion on the nature of linguistic rules.

The frequency of words in actual language use correlates with their

activation level in the mental lexicon. Statistical and probabilistic data

are therefore relevant for adequate models of morphological knowledge.

Questions

1. The list of differences between a dictionary and the mental lexicon insection 10.1 is not exhaustive. What other differences can you mention?


2. The following compounds coined by children have been observed by Elbers(1988): car-milk ‘‘gasoline’’,moon-nuts ‘‘cashew nuts’’. Why would childrencoin these words instead of using the existing words for these concepts?

3. Consider the data in the table concerning the frequency of words with theprefix un- and their bases in the CELEX database.

Which of these un-words do you expect to be processed via the directroute? To what extent does this correlate with your intuitions about thesemantic transparency of these words?

4. The following cases of noun–verb conversion occurred in the speech of mychildren (native speakers of Dutch) at the age of 2 to 4 (none of these verbsare used by adult speakers):

noun verbau ‘‘ouch’’ au ‘‘to hurt’’telefoon ‘‘telephone’’ telefoon ‘‘to telephone’’drop ‘‘liquorice’’ drop ‘‘to eat liquorice’’viltstift ‘‘felt pen’’ viltstift ‘‘to use a felt pen’’

How does this kind of word-formation relate to the principles of transpar-ency, simplicity, and conventionality that Clark (1993) identifies as playing arole in morphological acquisition?

5. Although to ride is a verb with the irregular past tense from rode, the pasttense of to joyride is joyrided. Try to explain this. How can this observation beused to defend a rule-based approach to morphology?

6. Consider the following speech errors

The speaker received a standing ovulation (ovation)His initiatives were engraved on the cigarette case (initials)A visual circle (vicious)

What conclusion can be drawn from these speech errors about how wordsare related in the mental lexicon?

7. Consider the data in the table concerning reaction times (in milliseconds)for a number of Italian nouns. Sometimes the singular form has the shortestreaction time, in other cases the plural form. How can this be explained?

Frequency Base frequency

uncanny 89 20unleash 65 16unscrew 44 187unaffected 54 169unobtrusive 42 17uncouth 34 2unkind 72 390

Source: Hay 2001: 1047.


8. One form of speech errors is blending two words. The following blendscan be found in Fromkin (1973) and are cited as evidence for productionmodels in Levelt (1989):

The competition is a little stougher [stiffer/tougher]Irvie is quite clear [close/near]At the end of today’s lection [lecture/lesson]

What kind of relationship between words in the mental lexicon must bethe basis for this kind of speech error?

9. Consider the following speech error in which two verbs have beenexchanged: . . . that I’d hear one if I knew it instead of the correct . . . thatI’d know one if I heard it. In Janssen et al. (2002) this observation is usedto argue that in language production first lexemes (such as KNOW

and HEAR) in their abstract form are selected before selection of theactual morphological forms takes place. Why does this speech errorsupport the assumption of these two stages in language production?

10. In the course of the history of Dutch, a number of Dutch verbs with a stemvowel ij [Ei] changed from regular verbs, with a suffixed past tense form, toablauting verbs, with a past tense vowel ee [e:]. This applies to, forinstance, the Dutch verbs prijzen ‘‘to praise’’ and lijken ‘‘to resemble’’,with the past tense stems prees and leek respectively (see Table 6.6 for asurvey of Dutch ablauting verbs). Note that the English counterparts ofthese verbs are regular. Other examples of such verbs that changedfrom weak to strong are schrijven ‘‘to write’’, belijden ‘‘to confess’’,and zwijgen ‘‘to keep silent’’. The class of ij-ee verbs comprises about aquarter of all ablauting verbs of Dutch. What does this change tell us aboutthe psychological reality of the vowel alternation patterns of irregularverbs?

Singular form Plural form Gloss

albero 583 alberi 542 treecapello 578 capelli 522 hairdente 527 denti 503 toothpiede 528 piedi 494 footsoldato 548 soldati 549 soldiernaso 523 nasi 556 nosepiazza 534 piazze 522 squaresorella 569 sorelle 534 sister

Source: Baayen, Burani, and Schreuder 1997: 30.


Further reading

A textbook completely devoted to the mental lexicon is Aitchison (1987). Clark(1993) presents an analysis and survey of studies on the acquisition of morph-ology, in particular word-formation. The importance of speech errors as lin-guistic evidence is shown in Fromkin (1973) and Cutler (1982).An extensive discussion on the issue of storage versus computation in rela-

tion to regular and irregular morphology can be found in Pinker (1999) forEnglish, and in Clahsen (1999) for German. Both defend a rule-based approachto morphology (with association patterns for irregular verbs only), and Pinkerargues in detail against the connectionist model. The ‘variable’ criterionfor calling something a rule can be found in Marcus (2001), and in Berentet al. (2002). Clahsen (1999) is followed by a number of invited criticalcomments.Evidence for the lexical storage of inflected forms is given in Baayen, Burani,

and Schreuder (1997) and in Baayen et al. (2002). A broader discussion ofthe issue of storage and computation for different domains of grammar canbe found in Nooteboom et al. (2002). Gurel (1999) argues that even for alanguage with rich agglutinative morphology such as Turkish, frequent wordsare accessed as whole words rather than through decomposition.Morphological race models are proposed in Frauenfelder and Schreuder

(1992), Baayen (1993), and Schreuder and Baayen (1997). The role of morph-ology in language production is discussed in Levelt (1989); Janssen et al.(2002) deals with inflectional forms in language production.A survey of the role of probabilistic data in morphology is presented in

Baayen (2003).


11Morphology and language

change

11.1 The historical perspective

Speakers are able to use their native language without any knowledge of

its history. Therefore, the historical perspective on language seems to

be superXuous when we want to understand how language works. In this

chapter it will be shown that this is not true: language change is a relevant

empirical domain for linguists who want to develop an adequate account of

language systems and their use.

First, we may want to know how the morphology of a language got the

properties that it has. Why do Germanic languages obey the Right-hand

Head Rule? And why did Dutch develop a system of linking elements

between the constituents of compounds? Such questions can only be

answered by looking at the history of languages.

A second reason for studying language change is that it provides another

window on the human mind. Patterns of language change tell us

how language users interpret the language data on the basis of which

11.1 The historical perspective 255

11.2 The nature of language

change 256

11.3 Historical sources of

morphology 261

11.4 Changes in morphological

rules 268

11.5 Changes in word

structure 272

Summary 275

Questions 276

Further reading 277

they develop their linguistic competence. Thus, they provide evidence

about what goes on in the minds of language users, in addition to what

psycholinguistic data can tell us.

In order to understand morphological change, it is useful to Wrst give

some attention to the nature of language change (section 11.2). We are then

ready to understand how morphological systems could develop (section

11.3) and how these systems and the complex words they create can change

(sections 11.4 and 11.5).

11.2 The nature of language change

The opening lines of GeoVrey Chaucer’s famous Canterbury Tales clearly

show that English has changed in the course of time:

(1) Whan that Aprill with his shoures soote

‘‘When that April with his showers sweet’’

The droghte of March hath perced to the roote

‘‘The drought of March has pierced to the root’’

And bathed every veyne in swich licour

‘‘And bathed every vein in such moisture’’

Of which vertu engendred is the Xour;

‘‘Of which virtue engendered is the Xower’’

Middle English as used in these lines is diVerent fromModern English, and

we might need some guidance in decoding these lines. If we assume that it is

still the same language, English, we have to conclude that languages can

change. But why do they?

The statement that languages change is in fact metaphorical in nature. It

presupposes that we conceive of a language as an organism that grows,

changes, and sometimes dies. This way of speaking suggests that languages

have amode of existence outside their users. This is true to a certain extent, as

we will see below, but a language primarily exists in the minds of its speakers.

That is why we say that a language has died when its last speaker has died,

as has, alas, happened often in recent years. So it is speakers that change

their language while using it in language perception and production.

Whydolanguageusers changetheir language? In fact, ‘change’ isnotalways

the appropriate word for what is going on. When a language acquires new

words through the activities of its users, it would be better to speak of con-


struction or innovation of language. What does change when new words or

newmeaningsofwordsget established is the lexicalnormof that language,not

the system behind it. The main reason for changing the lexical norm of a

language is that language users need expressions for new concepts, or new

things. One way of meeting this need is extending the meaning of existing

words. For instance, the wordmouse acquired a newmeaning as a navigation

device for computers. Alternatively, we may coin new denoting expressions,

either phrases or words. The phrase prosodic word has been introduced into

English to express a speciWc linguistic notion.Theword anti-virus-program is a

complex word that Wlls a lexical gap in the domain of computer technology.

Lexical gaps can also be Wlled through borrowings. Dutch, for example, has

borrowed many words from English recently. The word printer is a good

example: even though Dutch has the word drukker for ‘‘printer’’ (interpreted

asahumanagent), it uses theEnglishwordprinter for aprintingdevice.Onthe

other hand, German usesDrucker for both the human agent and the printing

device. Speakers of languagesmaydiVer in their attitudes towards borrowing.

If they try to avoid it, they are called purists.

The change of the lexical norm consists of two steps. The Wrst step is that

an individual language user coins a new lexical unit (a word or a phrase),

borrows one from another language, or gives an existing word a new inter-

pretation. The following step is that this innovation is accepted by other

language users, and is also used by them. This is also why dictionary makers

will not simply list in their dictionary each new word they come across in a

corpus of language use. It must have been used by diVerent users on diVerent

occasions before it is evidence for a change of the supra-individual lexical

norm. So we see that the kind of language change we may call lexical

innovation arises because language users need to express new concepts.

Maybe we should not call it language change, but language construction.

Although language is primarily located in each individual language user,

‘norm’ as introducedhere implies that it also has aderivative formof existence

at the supra-individual level, just likemanyother non-material objects such as

norms,knowledge,religions,andphilosophical ideas.Normsarethingswecan

share with other people. Sharing our language with other people is part and

parcel of being a language user. This paradox of language being simultan-

eously individual and social in nature has been stressed by Ferdinand de

Saussure (Saussure 1916), and is therefore called the Saussurean paradox.

The study of language change therefore has a psychological and a social

dimension. It will arise in the mind of one or more individual language

morphology and language change 257

users, but must Wnd its way in the language community before the new use

of a language form is actuated as a language change.

A second source of linguistic innovation besides changing the lexical

norm is reanalysis. Language users cannot grasp the system behind a

language in a direct fashion. The only evidence they have are outputs of

the system, concrete cases of language use. This opens up the possibility that

a language user reconstructs the system underlying the perceived outputs in

a slightly diVerent way from previous users. Some linguists have therefore

hypothesized that imperfect learning by children is the driving force behind

language change. Children have to reconstruct the rule system on the basis

of what they hear, and might thus come up with a slightly diVerent system

from the system of adult speakers that underlies these outputs. A morpho-

logical example is the potential regularization of past tense formation of the

irregular verbs of English discussed in Chapter 10. Children might start

using goed and bringed instead of went and brought. This kind of change is

impeded by the eVect of conventionality: children appear to adjust them-

selves quite readily to the adult system with its exceptions. Moreover, adult

speakers may also change their language through reanalysis, since they are

continuously interpreting the outputs that they perceive. In sum, language

users may be involved in language change at diVerent stages of their lives.

The types of change discussed so far are examples of internal change,

caused by the internal dynamics of language usage and by its transfer to the

next generation of language users. Language change may also be eVected

by external causes, in particular language contact. This is called external

change. Language contact may lead to borrowing of words from other

languages. Most Germanic languages borrowed many words from French,

for some centuries the dominant European language in the realm of cul-

ture, science, and diplomacy. English was inXuenced much earlier and more

strongly by French due to the Norman conquest of 1066.

Borrowing from French had its impact on the morphology of Germanic

languages. In Dutch, for instance, quite a number of aYxes of French

origin are used productively in combination with non-native stems, and

sometimes even with native stems (as in Xauw-iteit ‘‘silly joke’’ derived from

the native adjective Xauw ‘‘silly’’). These non-native aYxes also occur in

complex words that have no counterpart in French, a clear indication that

they have become aYxes of the borrowing language. That is, these aYxes

might have entered the language as parts of borrowed words, but are

recognized as morphological building blocks of these words. An example


of this kind of borrowing in English is the de-adjectival suYx -ity, like

Dutch -iteit the counterpart of French -ite. Once English had a number of

adapted loan words of the type X-ity, this pattern could be extended to

other adjectives, resulting in new nouns of this type.

Borrowing of inXectional forms may also take place. English has Greek

and Latin plurals, such as criteria, schemata, musea, and data. These learned

plurals came to be replacedwith regular plurals such as schemas andmuseums

(musea is no longer used in English). They are not always felt as plural

forms—witness the increasing tendency to use data as a subject NP with a

singular verbal form (The data is reliable,Much data was lost). InDutch, with

the sameborrowedplural formmusea, one often comes across the plural form

musea-s, which indicates that the stored form musea is no longer felt as an

unambiguous plural form. Similarly, one often hears data-s as the plural of

datum ‘‘date’’ instead of data or datums.

Lack or rarity of morphology is considered by some linguists to be

typical for pidgin and creole languages. Pidgin languages arise when, in a

situation of language contact, there is no common language. This situation

obtained, for instance, on plantations in the Americas with African slaves:

the slaves did not speak English or Spanish, while the plantation owners

did not speak the languages of the slaves. Creole languages are pidgin

languages that have come to be used as a Wrst language by a new generation

of language users. From recent research it has become clear, however,

that both pidgins and creoles do have morphology, both inXection and

word-formation (derivation, compounding, reduplication). For instance,

Haitian Creole borrowed many complex words from French, which

resulted in this language having productive derivational morphology

(Lefebvre 2003).

Language contact as a source of loss of inXectional morphology can be

observed in the development of Afrikaans. In South Africa, Dutch came

into contact with Portuguese, Malay, and local African languages, and thus

Afrikaans exhibits the language contact eVects that are typical for pidgi-

nization and creolization. Compared to its mother language Dutch,

Afrikaans has an extremely simple kind of inXectional morphology, with

a present tense form that is identical to the stem, without person and

number marking, and a periphrastic form for the non-present tense.

Likewise, the strong inXectional erosion of English during an earlier

episode in history may also be the eVect of language contact, namely that

between the original population of England and the Viking invaders.


A pervasive type of morphological change is the erosion of inXectional

systems in the Germanic and Romance languages, referred to as deXection.

English has the poorest inXectional system of all. Most Germanic languages

have lost their case system (German and Icelandic are exceptions), and the

three-gender system, again preserved in German and Icelandic, has been

reduced to two genders in Dutch and most North Germanic languages,

whereas English has lost gender in full NPs completely. This kind of

long-term change over a number of centuries has been dubbed drift by the

American linguist Sapir.

It is no coincidence that case and gender are the morphological cate-

gories that are subject to erosion. For instance, the distinction between

nominative and accusative case is a matter of contextual inXection,

required by syntactic context. This stands in opposition to inherent inXec-

tion such as number inXection for nouns, which expresses a semantic

distinction. The role of contextual inXection can be taken over by other

means of expression such as the use of prepositions or postpositions and

word order. Similarly, gender is a category that manifests itself in agreement

patterns, and thus plays a role in contextual inXection only. Thus,

although we cannot predict when and how languages change, it is possible

to identify linguistic factors that constrain and shape language change, such

as the distinction between inherent and contextual inXection.

Whatever the cause of language change, it is remarkable that language

users seem to cooperate in achieving a particular result such as the loss of a

language’s case system. In order to understand this, we might invoke the

invisible hand theory made famous by Adam Smith in his The Wealth of

Nations to explain how the economic market seems to behave as a rational

person, as if led by an invisible hand. Each individual action is a contribu-

tion to a result that has not been planned by individuals. If there are too

many cars on the road you have to use your brake from time to time, and all

drivers will do so. Quite soon, you will thus get stuck in a traYc jam,

although none of the individual drivers had the aim of creating it. In the

sameway, the acts of individual language usersmay have unforeseen eVects.

Language change is obviouslynot restricted to themorphological systemof

languages. Thequotation above fromChaucer, when compared to itsmodern

English glosses, also shows phonological, syntactic, and semantic changes.

Phonological change is exempliWed by the loss of the Wnal vowel of roote that

hasbecome root inModernEnglish.Thewordordernoun–adjective inshoures

soote correspondswith theModernEnglishwordorder adjective–noun: sweet


showers. An example of semantic change is that the present-day meaning of

liquor is no longer ‘‘liquid’’, but denotes alcoholic liquids only. For obvious

reasons, this chapter will focus on morphological change.

11.3 Historical sources of morphology

Morphological systems not only erode, they may also arise. An example is

the emergence of nominal compounding in Germanic languages that arose

from noun phrases of the type:

(2) [N-gen N]NP

In this type of noun phrase, the pre-nominal noun with genitive case

functions as a modiWer of the head noun. Such noun phrases were reinter-

preted as compound nouns. This emergence of Germanic compound struc-

ture is still reXected by the fact that sometimes an old case ending appears

at the end of the Wrst constituent. These case endings might be reanalysed as

allomorphic extensions of the Wrst constituent, or as linking elements. This

also applies to case endings such as the genitive case endings -s and -en of

Middle Dutch. Hence, we see transitions of the following kind:

(3) (des) her-en huis > her-en-huis

(the-gen) lord-gen house.nom lord-linking element-house ‘‘mansion’’

(des) konink-s krone > koning-s-kroon

(the-gen) king-gen crown.nom king-linking element-crown

‘‘royal crown’’

These changes thus led to the emergence of a system of compounding with

internal linking elements.

Univerbation, the reinterpretation of phrases as words, does not neces-

sarily lead to language change. Consider the following examples of uni-

verbation:

(4) Old High German hiu tagu ‘‘this day-instr’’ > Modern High German heute

‘‘today’’

Latin ad ipsum ‘‘to itself-acc’’ > Italian adesso ‘‘now’’

French peut etre ‘‘can be’’ > French peut-etre ‘‘perhaps’’

These changes imply the addition of lexemes to the lexicon, but do

not aVect the morphological system. That also applies to the following

conjunctions of Dutch:


(5) door-dat ‘‘due to’’, om-dat ‘‘because’’, op-dat ‘‘in order to’’

Such words arose from univerbation of a preposition with the Wrst word of

a clausal complement, the complementizer dat ‘‘that’’:

(6) Rebecca viel [[om]P [dat ze niets zag]S]PP

‘‘Rebecca fell for [that she nothing saw]S’’ reanalysed as:

Rebecca viel [omdat ze niets zag]S

Rebecca fell [because she nothing saw]S

‘‘Rebecca fell because she saw nothing’’

Although this kind of univerbation did not lead to new morphological

rules, there is an important lesson that we can draw from these historical

facts: when a word is multimorphemic, this does not necessarily imply that

it has been created by a morphological process. So the Dutch complemen-

tizers in (5) are not the products of compounding (combining prepositions

with complementizers), and we can thus maintain the regularity that word-

formation processes create words of lexical categories, that is, content

words only.

Univerbation did lead to a morphological rule in the case of Latin phrases

such as claramente ‘‘with a clearmind, abl’’. After univerbation and semantic

reinterpretation of this phrase, words such as claramente ‘‘in a clear manner’’

served as models for new de-adjectival words in -mente. Thus, -mente became

a productive suYx in Romance languages. This example illustrates that

twomechanisms are involved inmorphological change of this type: reanalysis

and analogy. The Wrst step is reanalysis: the existing word combination is

given another structural interpretation. The second step, which aVects the

morphological system, is the extension of the class of words in -mente through

analogy. The Wrst step pertains to the syntagmatic axis of language structure,

the second step to its paradigmatic axis. The compounds in (3) also show that

univerbation can lead to a morphological change, the emergence of nominal

compounding. The univerbated phrases were reinterpreted as compounds,

and new compounds were created directly on the model of the univerbated

ones, without having passed a phrasal stage.

An example of a productive preWx that derives from a syntactic con-

struction is the Frisian preWx witte- ‘‘very’’ that is a reduced form of the

word sequence wa wit hoe ‘‘who knows how’’. This word sequence forms a

clause in combination with an adjective, as inWa wit hoe Wer ‘‘Who knows

how far’’. This preWx is used productively in modern Frisian, as in the

following words (Dyk 1988).


(7) witte-Wer ‘‘very far’’

witte-heech ‘‘very high’’

witte-lang ‘‘very long’’

witte-djip ‘‘very deep’’

The speciWc meaning of the original word sequence underlying witte- has

changed into the more general meaning ‘‘very’’.

The emergence of this preWx illustrates that frequency plays a role in

grammaticalization, which has also been deWned as ‘the process by which a

frequently used sequence of words or morphemes become automated as a

single processing unit’ (Bybee 2003: 603). The word sequence wa wit hoe

expressing the meaning ‘‘high degree’’ is a clear example of such an auto-

mated, single processing unit. It is well known that high frequency triggers

phonological reduction: the speaker can aVord to be sloppy in the pro-

nunciation of a unit if that unit is easily retrievable for the hearer due to its

high frequency. Reduction will then lead to simpler phonological forms.

The reduced form witte [w1t

e

] has the phonological shape of a trochee.

Recall that a trochee is a foot that consists of two syllables, of which the

Wrst carries stress. In Germanic languages, the second syllable of a trochee

prefers to have schwa as its vowel. As we saw in section 7.2, Germanic

preWxes tend to be prosodic words of their own. An optimal prosodic word

consists minimally of a trochaic foot. Thus, the phonological reduction of the

clause wa wit hoe led to a preWx witte- with an optimal prosodic form.

Morphological processes not only arise through desyntactization, as was

the case for compounding, but also through dephonologization. A famous

case is that of umlaut in German. This originally phonological rule had the

eVect of fronting back vowels of stems when followed by a /i/ or /j/ in the

next syllable. In Early Old High German the declension for the word gast

‘‘guest’’ was as in (8). In later Old High German the instrumental form

changed into gastiu, thus creating a systematic correlation between the

(8) sg pl

nom gast gesti

gen gastes gestio

dat gaste gestim

acc gast gesti

inst gestiu

Source: Wurzel 1980: 448.


opposition singular : plural and the opposition back vowel : front vowel.

This kind of change in which formal diVerences within the same (sub)par-

adigm are removed is called paradigmatic levelling. In Middle High Ger-

man, the case ending vowel reduced to schwa, resulting in the following

paradigm (the separate instrumental case had disappeared). When the

(9) sg pl

nom gast geste

gen gastes geste

dat gaste gesten

acc gast geste

Source: Wurzel 1980: 448.

syllables with these sounds were reduced to syllables with schwa, the

conditioning environment for the vowel change (the presence of a high

vowel or glide) was destroyed. Yet, the contrast between back and front

vowels was maintained, and this vowel alternation thus obtained the

function of marking the plural, as in modern German Gast–Gaste [gAst–gEst

e] (with the diaeresis still representing the umlaut in the orthography).

In this example, the plural is also marked by the ending -e, but in other

cases the plural is expressed by umlaut only, as in:

(10) Apfel–Apfel ‘‘apple(s)’’

Kloster–Kloster ‘‘convent(s)’’

Mutter–Mutter ‘‘mother(s)’’

Vater–Vater ‘‘father(s)’’

Umlaut may co-occur with the overt plural suYxes -e and -er, and also

occurs in complex words with certain derivational suYxes such as -lich (as

in Vater ‘‘father’’–vaterlich ‘‘fatherly’’).

Umlaut might be analysed in two ways. One possibility is claiming that

umlaut has completely morphologized. This means that it has become

similar to ablaut in that it expresses a morphological property (plural) by

means of vowel change. This analysis implies that the property plural is

sometimes marked doubly, through vowel alternation and suYxation with

-e, as in Gast–Gast-e ‘‘guest sg/pl’’. The other option is interpreting umlaut

as a case of stem allomorphy triggered by particular morphological cat-

egories such as plural and words with the adjectival suYx -lich. In that


analysis, the property plural is also marked twice in Gaste, through stem

allomorphy and suYxation.

This umlaut system has been preserved remarkably well in modern

German, whereas it has almost completely disappeared in English, with a

few alternations such as foot–feet and goose–geese being the only remnants.

It is interesting to see how the reduction of allomorphy in a paradigm

(that is, paradigmatic levelling) applies, and which allomorph is kept. In

singular–plural pairs of nouns, it is usually the stem-form of the singular

that is kept. However, if the plural form is more frequent than the

singular form, it is the plural stem-form that may win. In Frisian there is

an allomorphy pattern called breaking. An unbroken (falling) diphthong in

the singular stem alternates with a broken (rising) diphthong in the plural

stem. The pattern of paradigmatic levelling in (11) has been found in

Frisian (Tiersma 1982: 834):

(11) Conservative dialect Innovative dialect

a. hoer [hu

e

r]–hworren [hwor

e

n] hoer [hu

e

r]–hoeren [hu

e

r

e

n] ‘‘whore’’

koal [ko

e

l]–kwallen [kwAl en] koal [ko

e

l]–koalen [ko

e

l

e

n] ‘‘coal’’

poel [pu

e

l]–pwollen [pwol

e

n] poel [pu

e

l]–poelen [pu

e

l

e

n] ‘‘pool’’

b. earm [ie

rm]–jermen [jErme

n] jerm [jErm]–jermen [jErme

n] ‘‘arm’’

toarn [to

e

rn]–twarnen [twArn e

n] twar [twArn]–twarnen [twArn e

n]

‘‘thorn’’

trien [tri

e

n]–trijinnen [trjin

e

n] trjin [trjin]–trjinnen [trjin

e

n] ‘‘tear’’

In (11a), the singular forms are more frequent than the plural forms, hence

levelling goes into the direction of the singular form. In (11b), on the other

hand, the plurals are used more often than the singulars because they

denote entities that come in a pair (arms), or in groups (thorns, tears).

This led to levelling in the direction of the plural allomorph. The levelling

of Frisian breaking is thus a case of bidirectional levelling.

An important source of morphology is the phenomenon of grammati-

calization. This phenomenon is usually deWned as follows:

(12) ‘[G]rammaticalization is [ . . . ] that subset of linguistic changes throughwhich a

lexical item in certain uses becomes a grammatical item, or through which

agrammatical itembecomesmoregrammatical’ (HopperandTraugott1993:2).

An example of the change from lexical to grammatical item is the develop-

ment of verbs into auxiliaries. In English the verb to have not only functions

as a main verb, with the meaning ‘‘to possess’’, but also as an auxiliary

in perfect tense forms. The verb can has lost its status as lexical verb


completely, and functions as a modal auxiliary only. In these examples

grammaticalization does not create morphology. This does happen if a

(lexical or grammatical) morpheme becomes an aYx. Dutch has a product-

ive preWx door- exempliWed in (13):

(13) door-boor ‘‘to drill through completely’’

door-ploeg ‘‘to through plough, to plough completely’’

door-zien ‘‘to through-see, to grasp’’

This preWx arose through reinterpretation of the directional postposition

door ‘‘through’’. For instance, the rise of the verb doorboor can be pictured

as follows:

(14) [[de muur]NP [door]P]PP [boor]V > [de muur]NP [door[boor]V]V

the wall through drill

‘‘pierce the wall’’

This is another example of reanalysis: a structural interpretation is assigned

to a sequence of words that diVers from the original one. The second

ingredient is again analogy: the reanalysed pattern must be extended to

other cases in which the original structure does not play a role. This is

indeed the case here: door- has become a productive preWx in Dutch.

Grammaticalization often goes together with the semantic development

of bleaching: the meaning of the grammaticalized element becomes more

abstract. The meaning of the Frisian preWx witte- discussed above is a clear

illustration of bleaching. In the case of door-, its originally concrete spatial

or locative meaning has developed into a more abstract aspectual one.

It indicates the completeness of the action performed: the object is

completely aVected.

A classic example of the development of a suYx out of a word is the past

tense suYx in Indo-European. In many Indo-European languages it con-

tains a dental or alveolar consonant (hence it is called the dental preterite).

This dental preterite developed from the proto-Indo-European verb for do

which already had this dental consonant. In Romance languages, some of

the inXectional endings in verbal paradigms arose from grammaticalization

of the Latin verb habere ‘‘to have’’. This is the case for French future forms

that arose from a periphrastic Latin form, as illustrated in (15) (Hopper

and Traugott 1993: 10):

(15) cantare habemus > (nous) chanter-ons ‘‘we will sing’’

The ending -ons which derives from habemus combines with a stem form

which is identical to the inWnitive form chanter, the French counterpart of


the Latin inWnitive form cantare. The choice of stem form thus reXects the

history of the future inXectional form.

A lexical morpheme can also grammaticalize when it forms part of a

complex word. The non-head constituents of compounds, which function

as modiWers, often develop into preWxes with a more general interpretation.

Examples are the Dutch lexical morphemes bloed ‘‘blood’’ and kut ‘‘cunt’’:

the Wrst has developed into a general intensifying preWx, the second into a

general preWx of negative evaluation:

(16) a. bloed-heet ‘‘lit. blood-hot, very hot’’

bloed-mooi ‘‘lit. blood-beautiful, very beautiful’’

bloed-link ‘‘lit. blood-dangerous, very dangerous’’

b. kut-wijf ‘‘lit. cunt-wife, nasty woman’’

kut-auto ‘‘lit. cunt-car, bad car’’

kut-smoes ‘‘lit. cunt-excuse, bad excuse’’

The same can happen to the heads of compounds. The Dutch suYxes

discussed in section 7.2 that were qualiWed as non-cohering derive historically

from lexical morphemes. For instance, the non-cohering suYx -baar derives

from the verb stem baar ‘‘to carry’’, and -heid (English -hood ) derives from

the earlyGermanicword heid ‘‘state’’. English suYxes such as -dom and -hood

as in freedom and childhood derive from nouns. That is, words containing

these morphemes used to be compounds, but their head nouns have been

reanalysed as suYxes since they donot occur anymore aswords on their own.

The new life of the morpheme -baar as a suYx was enhanced by another

type of reanalysis. As we have seen, aYxes may impose constraints on the

word class of their bases. Originally, the lexical morpheme baar combined

with nouns, as in German frucht-bar, Dutch vrucht-baar ‘‘fertile’’, both

originally compounds meaning ‘‘fruit-carrying’’. In some words in -baar,

the Wrst constituent could be reinterpreted as a verb, as in strijd-baar ‘‘mili-

tant’’. The base word strijd is both a noun ‘‘Wght’’ and a verb ‘‘to Wght’’.

Subsequently, -baar could be reinterpreted as a suYx that takes verbal bases,

and it is in this use that it has become productive in Dutch and German.

The distinction between aYx and lexical morpheme tends to be blurred in

the case of cliticization, because both aYxes and clitics require a host word

to combine with. In Finnish and Hungarian, some case endings arose from

cliticized postpositions. This ambiguity of cliticized morphemes also ex-

plains why in a number of Germanic languages the genitive case ending -s of

nouns in pre-nominal position could be reinterpreted as a phrasal aYx, that


is a clitic. This latter, reverse development is called degrammaticalization

since a suYx has developed into a grammatically independent possessor

marker. This is the case for s in phrases such as the king of England’s hat.

The phenomenon of degrammaticalization has raised the question

whether grammaticalization is subject to the condition of unidirectionality.

Is there only one irreversible direction of change in grammaticalization,

from lexical to grammatical, and from grammatical to more grammatical

morphemes, or can a grammatical morpheme become less grammatical, or

even lexical as well? This depends on one’s deWnition of grammaticaliza-

tion. If unidirectionality is one of its deWning properties, changes into the

opposite direction simply require another descriptive term, for instance

degrammaticalization. The reinterpretation of the English genitive case

morpheme -s as a clitic is a restricted form of degrammaticalization. It is

a restricted case because the suYx does not end up as a full lexical word. In

Early Modern English it has also been interpreted as the weak form of the

possessive pronoun his, that is, as a full word. Thus, Shakespeare writes the

count his galleys ‘‘the count’s galleys’’ (Greg Stump, personal communica-

tion). This change is related to the fact that English lost its case system

(except for pronouns), and this made a reanalysis of the genitive suYx -s

necessary. There appears to be consensus that degrammaticalization is

restricted in nature, whereas grammaticalization is a pervasive phenom-

enon in natural languages.

11.4 Changes in morphological rules

Erosion of inXectional morphology has taken place in most Germanic and

Romance languages. This means that some inXectional rules have disap-

peared from the grammar of those languages. Dutch, for instance, has lost

its morphological cases. The case system has only survived in a number of

Wxed expressions, as illustrated by the following examples, prepositional

phrases in origin. The prepositions in these phrases govern a particular

case. Some of these frozen expressions are considered as words, others are

still spelt as phrases. The case endings are in italics:

(17) te-gelijker-tijd ‘‘at the same time, simultaneously’’

met dien verstande ‘‘with that understanding’’

in dier voege ‘‘in such a manner’’


An interesting case of change in inXectional morphology is that of

adjectival inXection in Afrikaans. In Dutch, adjectives in attributive

position are inXected according to the following rule: ‘add the suYx -e [

e

]

to the adjectival stem unless the NP in which it occurs carries the features

[indeWnite], [neuter], and [singular]’. Thus we get the following contrasts

for NPs with attributive adjectives (paard ‘‘horse’’ is neuter, koe ‘‘cow’’ is

non-neuter):

(18) een groot-ø paard ‘‘a big horse’’ een grot-e koe ‘‘a big cow’’

het grot-e paard ‘‘the big horse’’ de grot-e koe ‘‘the big cow’’

(de) grot-e paarden ‘‘(the) big horses’’ (de) grot-e koeien ‘‘(the) big cows’’

(The variation groot/grot is due to spelling conventions.) The dependence

of the absence of the inXectional schwa on three diVerent features is quite a

complex system, witness the fact that many non-native speakers of Dutch

never learn to master it properly and always add -e to the prenominal

adjective. No wonder this system broke down in Afrikaans, a creolized

variant of Dutch. In the new system that developed, the adjectives are

divided into two classes: either they always carry the inXectional schwa in

attributive position, or they never do. The adjectives that do take the schwa

are morphologically complex adjectives, and simplex adjectives that exhibit

stem allomorphy. For instance, the adjective sag ‘‘soft’’ has the stem

allomorph sagt, as shown in the inXected form sagte. Hence, it will have

the form sagte in attributive position, whereas sag is used in predicative

position. What we can learn from these facts is that relics of a previous

stage of the language, ‘‘historical junk’’ in the words of Lass (1990), might

be kept, and even re-used in a diVerent way.

Afrikaans exhibits another kind of rule change, rule simpliWcation. Both

in Dutch and in Afrikaans, there are two competing plural suYxes: -s and

-en /

e

n/. The choice between these two suYxes is governed by a prosodic

constraint: -s is selected after an unstressed syllable, -en after a stressed one.

The eVect of this selection principle is that a plural noun always ends in a

trochee. In Dutch, this prosodic principle is overruled in some types of

complex words. For instance, de-adjectival nouns in -e /

e

/ always select

the plural ending -en /

e

n/ (with concomitant prevocalic schwa-deletion),

although -swould also be in line with the prosodic constraint. In Afrikaans,

this morphological factor is ignored, and the plural suYx for complex

nouns is selected in the same way as for simplex nouns. Thus -s is selected

for words ending in the suYx -e (Van Marle 1978: 147):


(19) noun Dutch plural Afrikaans plural

blank-e ‘‘white person’’ blank-e-n blank-e-s

geleerd-e ‘‘scholar’’ geleerd-e-n geleerd-e-s

gevangen-e ‘‘prisoner’’ gevangen-e-n gevangen-e-s

In the realm of derivational morphology two types of change are prom-

inent. First, rules may change as to their productivity. Dutch has two

suYxes for deriving deverbal adjectives, -baar ‘‘-able’’ and -(e) lijk

‘‘-able’’. In Middle Dutch the suYx -elijk was still productive. However,

in the standard variety of Modern Dutch it can no longer be used to coin

new adjectives, unlike its competitor -baar. Thus, we see that aYxes can

lose their productivity. Another example of a now unproductive suYx is

the Dutch suYx -el that we Wnd in nouns such as in (20).

(20) base derived noun

drup ‘‘drop’’ drupp-el ‘‘drop’’

eik ‘‘oak’’ eik-el ‘‘acorn’’

ijs ‘‘ice’’ ijz-el ‘‘freezing rain’’

This suYx derives from an Indo-European suYx -l that indicated descent,

and hence also diminution. Why has it become unproductive? An inspec-

tion of Dutch nouns in -el will reveal that for most of them there is no base

noun. This applies for example to the following nouns:

(21) korrel ‘‘grain, granule’’, kruimel ‘‘crumble’’, kwartel ‘‘quail’’, pukkel ‘‘pimple’’,

sleutel ‘‘key’’, stempel ‘‘stamp’’, vezel ‘‘Wbre’’, wezel ‘‘weasel’’

Consequently, these nouns will not be parsed into morphological subcon-

stituents, and thus the suYx -el will have an extremely low activation level.

That is, these words have lost their morphological structure.

A second observation on these nouns is that their semantics tend to vary:

they do not belong to one clear semantic category. Consequently, most

speakers of Dutch do not consider words in -el as complex. These obser-

vations support the view that the abstract patterns we call word-formation

rules depend for their existence on the formal and semantic transparency

of the complex words over which they generalize. The existence of trans-

parent complex words is necessary for a word-formation pattern to remain

productive.

Another type of possible change in derivational rules is a change in the

category of the base words to which they apply. Very productive rules tend

to extend their domain of application to new categories. The productive


diminutive suYx -tje of Dutch, for instance, is no longer restricted to the

domain of nouns, but also attaches to adjectives, verbs, and adverbs:

(22) strijk ‘‘to stroke’’ strijk-je ‘‘small string orchestra’’

blond ‘‘blond’’ blond-je ‘‘blond girl’’

uit ‘‘out’’ uit-je ‘‘outing’’

Category change of the inputs of a word-formation rule may occur as the

result of reinterpretation, as happened in the case of the suYx -baar, as

discussed above.

The vicissitudes of the French suYx -age also illustrate category change

(Fleischmann 1977). This suYx derives from the Greek–Latin adjectival

suYx -aticus. The neuter forms of these adjectives could be used as nouns,

as in viaticum ‘‘money for a journey’’. In addition, -agewas reinterpreted as

being nominal on the basis of the elliptical use of phrases. For example, the

French noun fromage ‘‘cheese’’ derives from the phrase caseu formaticu

‘‘cheese made in a mould’’ (with the adjective formaticu ‘‘moulded’’), and

thus changed its status from adjective to noun. Thus, -age could be inter-

preted as a nominalizing suYx. In Medieval French, this nominalizing -age

attached to nouns, as in mouton-age ‘‘tax on a lamb’’ and hommage ‘‘hon-

our’’ (from homme ‘‘vassal’’). This -age is also found in the French word

langage, and its English counterpart language. In Modern French, this

nominalizing suYx is no longer productive with nouns, but only with

verbs, as in lavage ‘‘washing’’ from laver ‘‘to wash’’. This change of base

category was probably brought about by reinterpretation. Some base

words could be interpreted as either noun or verb. For instance, aunage

‘‘measuring by the ell’’ could be seen as derived from either the noun aune

‘‘ell’’, or the verb auner ‘‘to measure by the ell’’.

The suYx -age entered many European languages through borrowing of

French words in -age. English has extended its use to native, Germanic

nominal and verbal stems, which shows its productivity in Modern

English:

(23) base N: foot-age, front-age, mile-age, shipp-age, wreck-age, yard-age

base V: break-age, brew-age, cover-age, drain-age, leak-age

The word shortage shows that -age has even been extended to adjectival

bases. In Dutch we Wnd both denominal and deverbal nouns in -age; as in

English, it has also been attached to native stems, an indication of its

productivity.


11.5 Changes in word structure

Complex words, once they are coined, may be subject to reanalysis

and reduction. In reanalysis, words receive a diVerent structural interpre-

tation. A classical example is the reinterpretation of the word hamburger.

This word, derived from the base noun Hamburg, and denoting a speciWc

kind of food originating from that city, received the following structural

reanalysis:

(24) [[hamburg]er] > [[ham][burger]]

Reanalysis can only be observed when the reanalysed structure serves as

a model for new words. In this case, many new words in -burger have been

coined, such as beefburger, cheeseburger, Wshburger, and the like. A similar

pattern is the coinage of turkeyfurter on the model of frankfurter. The piece

burger itself has been reinterpreted as a noun, witness the brand name

Burger King. So in fact we cannot conclude that a new suYx -burger has

developed in English. Burger is a new noun denoting a particular kind of

fast food, and can be used as the head of nominal compounds.

Semantic reinterpretation may also lead to a new set of words. This

has been the case for -gate, as used in Watergate (the name of a building

in Washington, DC, that was burglarized by order of President Nixon).

The morpheme -gate received a new interpretation, ‘‘political scandal’’,

thus leading to many new words such as Monicagate and Irangate (cf.

question 4.4). This suYx also acquired the more general meaning of

‘‘scandal’’, as in nipplegate, a scandal in which the pop star Janet Jackson

was involved. We may conclude that -gate has become a suYx because

this speciWc meaning of gate is only available when it is combined

with a noun. This suYx -gate has been borrowed massively in most

European languages. For instance, a recent political scandal around

the prospective bride Mabel of one of the Dutch princes is referred to

as Mabelgate.

Reanalysis may also have the eVect of a sequence of aYxes becoming an

aYx. A morphological structure [[[x]A]B] can be reinterpreted as [[x]AB].

An example from Dutch is the suYx -erij, originally a combination of the

suYxes -er and -ij:

(25) base noun in -er noun in -ij

bak ‘‘to bake’’ bakk-er ‘‘baker’’ bakk-er-ij ‘‘bakery’’ > bakk-erij


This suYx -erij is now productively attached to verbal bases without an

intermediate noun in -er being necessary), and also to base nouns, which

conWrms that -erij has started a life of its own:

(26) race ‘‘to race’’ race-erij ‘‘racing’’

drogist ‘‘chemist’’ drogist-erij ‘‘drugstore’’

This phenomenon of two aYxes becoming one is referred to as aYx

telescoping. It also occurred in French where bijouterie ‘‘jewellery’’ could

be reinterpreted as bijout-erie (bijou ‘‘jewel’’ > bijout-ier ‘‘jeweller’’ > bijout-

er-ie ‘‘jewellery’’). This French telescoped suYx -erie was also borrowed in

Middle Dutch, and it became -erij due to a regular sound change of

diphthongization of /i/ to /Ei/. Thus, the borrowed -erij enhanced the rise

of a native suYx -erij.

A type of morphological change that concerns individual words is

systematization. In Turkish, many nouns denoting persons end in the

suYx -cI (the capital I stands for the set of high vowels, because the vowel

of this suYx is subject to vowel harmony). This suYx may also be found

attached to loan words from French such as the following (Lewis 1967: 60,

cited in Van Marle 1978: 148):

(27) sofor ‘‘chauVeur’’ sofor-cu

garson ‘‘waiter’’ garson-cu

The addition of this suYx to loan words has the eVect that all nouns

denoting persons are denoted in a uniform way. That is why this morpho-

logical change is qualiWed as systematization. This morphological pattern

complies with a tendency in natural language that is called the one mean-

ing—one form principle: each meaning should correspond with one form.

That is, sameness of meaning implies sameness of form. Compliance with

this principle is achieved in this case by adding a suYx that is strictly

speaking superXuous, and hence a pleonastic addition. Thus, in this case,

systematization leads to overcharacterization.

Such pleonastic aYxation is also found with Dutch acronyms that denote

certain professions or ranks. Quite often these acronyms are enriched with

the suYx -er which creates such denominal names (as in wetenschapper

‘‘scientist’’ from wetenschap ‘‘science’’):

(28) UD (< Universitair Docent ‘‘university teacher’’) > UD-er

KVV (< Kort Verband Vrijwilliger ‘‘short-term volunteer’’) > KVV-er


For the same reason, some complex words of Dutch ending in schwa were

adapted, with replacement of -e by -er, and personal names were extended

with -er, as in:

(29) a. herd-e ‘‘shepherd’’ > herd-er, scutte ‘‘shooter, riXeman’’ > schutt-er,

schenke ‘‘cupbearer’’ > schenk-er;

b. Dominic-aan ‘‘Dominican’’ > Dominic-an-er, Farizee ‘‘Pharisee’’ >

Farizee-er

An example of aYx substitution from child language that is also a case of

systematization is the replacement of -ig by -baar in the Dutch adjective

door-zicht-ig ‘‘lit. through-see-able, transparent’’, leading to door-zicht-

baar. It is the suYx -baar that normally expresses the meaning ‘‘-able’’,

and hence this adaptation by my daughter Suzanne at the age of 2 is also a

case of systematization.

Reduction of word-internal structure takes place when a complex word is

no longer semantically transparent. In that case, itmay lose itsmorphological

structure and change phonologically into the direction of the canonical

phonological form of simplex words. The Dutch word aardappel ‘‘potato’’

is originally a compound made from the lexemes aard ‘‘earth’’ and appel

‘‘apple’’. However, it is not felt as a kind of apple, and hence became opaque.

This reinterpretation is reXected by the way this word is parsed into syllables.

Whereas each lexical constituent of a compound is an independent domain of

syllabiWcation, the word aardappel is now parsed as a simplex word:

(30) compound [[aard]N[appel]N]N (aard)ø(ap.pel)ø [a:rtAp e

l] reanalysed as

simplex word [aardappel]N (aar.dap.pel)ø [a:rdAp e

l]

The diVerent patterns of syllabiWcation (indicated by the dots) which reXect

the loss of morphological structure have an eVect on the phonetic form of

this word. In Dutch, syllable-Wnal obstruents are devoiced. Hence, the

phonetic form [a:rdAp e

l] with a [d] betrays the loss of morphological

structure.

A second example is the Dutch adverb natuurlijk ‘‘of course’’ that derives

from the denominal adjective natuur-lijk ‘‘natural’’ (< natuur ‘‘nature’’). In

its adverbial use, there is no longer a clear semantic relation to its base noun

natuur, and hence this frequently used adverb is pronounced as tuurlijk

[ty:rl

e

k] in casual speech. The latter phonetic form has the shape of a

trochee, with the second syllable headed by a schwa, which is the optimal

prosodic form of Dutch simplex words. The only marked aspect of this


phonetic form is that the Wrst syllable contains a long vowel followed by a

consonant, whereas Dutch word-internal syllables prefer to be bimoraic

(that is, they end either in a long vowel or in a short vowel þ one conson-

ant). This constraint is obeyed in the even more reduced form of natuurlijk

that we also Wnd in casual speech: [tyl

e

k]. When used as an adjective,

however, the word natuurlijk has a transparent morphological structure,

and cannot be reduced phonetically.

PreWxes may also lose their morphological status. This is quite clear in

English borrowings from Latin such as abortion and adoption. These words

contain the preWxes ab- and ad- respectively. In transparent complex

words, a preWx boundary coincides with a syllable boundary. In these

words, however, this is not case: the syllabiWcations are a.bor.tion and

a.dop.tion respectively. The Dutch preWx ge-, as in geloof ‘‘to believe’’ is

no longer productive as a verbal preWx. In Afrikaans, the verb geloof has

become a simplex verb, glo, with deletion of the schwa (remember that

Dutch simplex words prefer to begin with a syllable containing a full vowel,

cf. section 7.2). This adaptation implies that in Afrikaans the past participle

is ge-glo, whereas in Dutch it is ge-loof-d, the regular form for verbs with an

initial unstressed preWx.

Summary

Word-formation processes lead to newwords, and hence the morphological

system of a language contributes to lexical innovation. The morphological

system itself is also a potential target of change, with external factors and

internal factors involved. Forms of external change are borrowing and

simpliWcation, both due to language contact. Borrowing of complex

words may lead to enrichment of the morphological system, whereas

simpliWcation means reduction of its complexity. Internal causes of change

are that each generation has to Wnd out the rules behind the system (imper-

fect learning, whichmay also lead to simpliWcation), and that language users

analyse the outputs of the linguistic system, which may lead to reanalysis.

Morphology may develop in the course of history from syntactic con-

structions through univerbation and grammaticalization. In addition,

phonologically conditioned allomorphy can be preserved after the loss of

the conditioning phonological environment, and may be reinterpreted as

marking morphological distinctions.


Complex words, once established, may undergo changes. They may get

a diVerent shape through systematization and overcharacterization.

Complex words may lose their semantic transparency, and hence their

morphological structure. This loss of structure can sometimes be inferred

from the way in which they are syllabiWed, or from phonetic reduction,

which makes them phonologically more similar to simplex words.

Questions

1. The existence of the English suffix -able is due to borrowing of Frenchadjectives in -able. Nowadays, it can also be attached to native verbalstems (doable, readable). What evidence can you provide for this suffixbehaving as a non-native suffix of English as well?

2. What is the relationship between bidirectional levelling and the findings onresponse latencies for the Italian nouns mentioned in question 10.7?

3. English has two plural forms for brother, brothers and the archaic brethren.These plural forms have different meanings in present-day English. Try toexplain why these words have different meanings.

4. The English word shepherd [sEp erd] derives historically from the compoundsheep-herd. Give an account of the phonetic reduction that this compoundunderwent, and why this reduction could take place.

5. The Italian word for ‘‘tomato’’ is pomodoro (plural pomodori ). Its originalform is pomo d’oro ‘‘apple of gold’’. Why can’t we assume this lexical unitis still phrasal in nature?

6. Try to find violations of the one meaning—one form principle in your nativelanguage.

7. The Latin word requiem is the ACC.SG form of the lexeme requies ‘‘rest’’, and isused in many languages to denote the Roman Catholic Mass for the Dead.How can it be explained that languages may apparently borrow cases ofcontextual inflection from other languages?

8. The Dutch compound scheidsrechter ‘‘referee’’ consists of the verbal stemscheid ‘‘to separate’’ and the noun rechter ‘‘judge’’. The linking elementbetween the two constituents is s. Why can’t we say that this s is a lexica-lized genitive case ending, and why must it be seen as a stem extension orlinking element?

9. Bakker (2003: 23) established the following hierarchy of preservation ofinflectional marking on nouns in pidgin languages: number > case> gender.This means that marking for number is more common than for case, andmarking for case is more common than for gender. How might this hier-archy of preservation be explained?


10. Compare the following two NPs with the same meaning from Old Swed-ish and Modern Swedish respectively (Norde 1997: 261):

Old Swedish: en-s man-s syndone-GEN man-GEN sin

Modern Swedish: en man-s syndone man-s sin‘‘one man’s sin’’

How can this change in the morphological marking of the possessor of thehead noun be classified?

Further reading

The invisible hand theory of language change is proposed in Keller (1994).Aitchison (2003) points out that children play a less prominent role in languagechange than sometimes claimed.The morphological erosion in pidgin and creole languages is discussed in

McWhorter (1998). As argued in detail in a number of articles in Plag (2003a),the idea that pidgin and creole languages do not have morphology is incor-rect.A good survey of the phenomena and the theoretical debates concerning

grammaticalization is found in Heine et al. (1991), Hopper and Closs Traugott(1993), Bybee et al. (1994), and Heine (2003). Heine and Kuteva (2002) is areference work on grammaticalization phenomena in the languages of theworld. Bybee (2003) stresses the role of frequency in grammaticalization.Bybee et al. (1994) and Closs Traugott (2003) argue that it is the reinterpret-ation of the whole construction of which a lexical word forms part that leads togrammaticalization of that lexical word. Good surveys and critical evaluationsof grammaticalization theory are given in Newmeyer (1998) and Campbell(2001). Norde (1997, 2001) discusses degrammaticalization. The rise of verbalparticles and prefixes in Germanic languages through reanalysis is discussed inmore detail in van Kemenade and Los (2003) and in Blom and Booij (2003).The remarkable persistence of paradigm irregularities in morphological

change is discussed in Maiden (1992).



Part VIConclusions


12The word as a linguistic unit

12.1 The notion ‘word’

The notion ‘word’ plays a central role in this book, as witnessed by its title,

The Grammar of Words. It is therefore useful to reXect on our use of this

notion, now that we have come to the end of this book, and should think

about what we have learned about the nature of this linguistic unit.

The basic approach that we have been following is that we started from

an intuitive notion ‘word’. That is, I assumed that you have some initial

understanding of what a word is. Indeed, it has often been observed in the

literature that the notion ‘word’ is one of the most accessible linguistic

notions for language users without linguistic training. This is particularly

the case for speakers of a language with an orthographical system. The

word in its orthographic sense in a language with an alphabetical script,

deWned as a sequence of letters not interrupted by spaces, is an accessible

unit. Yet, even such a simple notion raises problems when we try to apply

it. For instance, this deWnition would mean that the English compound

syllable boundary spelled with internal spacing should not be considered as

one word, but as a sequence of two words, which goes against the intuition

of speakers of English. How are we going to Wnd a principled answer to the

question of identifying and delimiting words?

12.1 The notion ‘word’ 281

12.2 The demarcation of words

and phrases 284

12.3 Phonological words 286

12.4 Lexical integrity 289

Summary 293

Further reading 293

A classical deWnition of the notion ‘word’ can be found in BloomWeld’s

Language that deWnes the word as a ‘‘minimum free form’’ (BloomWeld

1935: 178). This means that a word can form, at least in principle, a

linguistic utterance in isolation. For instance, when you ask What are we

going to eat today? the answer might be Lasagna. This proves that lasagna is

a free form, and since it cannot be decomposed into smaller meaningful

units, it is a minimum free form, and hence a word. You will not Wnd it

diYcult, however, to come up with cases where this simple criterion does

not work. You cannot answer the question What is that? with the word:

Apple. Instead, for reasons of syntax, you have to say: An apple, although

you will agree that apple is a word in English. What one also has to exclude

is metalinguistic usage, because in that case we can also use bound

morphemes in isolation (Which suYx appears in ability? -ity).

An additional and often used criterion for wordhood, aimed at distin-

guishing words from phrases is the criterion of cohesiveness or un-

interruptability: you cannot interrupt a word by putting other words in

between its constituents. For example, since we can say John looked the

information up, we should not consider look up as one word in the sentence

John looked up the information, although semantically look up does form a

unit.

Such deWnitions and criteria are a good starting point, but not the end-

point of a scientiWc analysis of the linguistic unit of the word. In this book,

I developed a more sophisticated notion ‘word’ in such a way that it

became a theoretical notion, embedded in a general theory of language

structure. This is a normal procedure in empirical sciences. One starts with

some intuitive notion, and subsequently, by observing phenomena, and

developing theoretical analyses of these phenomena, a corresponding the-

oretical notion will be Xeshed out.

A Wrst step in moving from an intuitive notion ‘word’ to a more sophis-

ticated one has been presented in Chapter 1, where a distinction was made

between ‘lexeme’ and ‘word form’. This is an essential distinction in using

the notion ‘word’. For instance, when one wants to determine, for whatever

reason, howmany diVerent words occur in a particular text, the answer will

depend on whether you mean ‘diVerent lexemes’ or ‘diVerent word forms’.

You will recall from Chapter 1 that the diVerence between these two

notions of ‘word’ presupposes the notion of inXection, and the idea that

word forms with diVerent inXectional markings can be seen as manifest-

ations of the same lexeme. This example thus illustrates an essential insight

282 conclusions

of the philosophy of science: the content of theoretical terms is determined

by the way in which they are embedded in a theoretical framework. Hence,

the theoretical distinction between ‘lexeme’ and ‘word form’ forms part

of the theoretical deWnition of the linguistic notion ‘word’.

A second important step in developing the theoretical notion ‘word’ is

the distinction between the notions ‘word’ and ‘lexical unit’, also intro-

duced in Chapter 1. It is well known that many language users tend to

identify these two notions. In particular, the idea is that a linguistic unit

that is given an entry in a dictionary is a word. We have seen, however, that

this identiWcation is wrong since it impedes a proper analysis of the way in

which linguistic units are structured and behave as parts of larger expres-

sions. That is why the notion ‘listeme’ is useful. Recall that a listeme is a

linguistic unit (a word or a multi-word unit) that is listed in the lexicon. It is

quite clear that language users are able to memorize chunks larger than

words in their lexical memory, such as phrasal terms (red tape, yellow fever,

etc.), idiomatic phrases and even complete sentences such as proverbs. As

we saw, our memory is vast, and can store a lot of information (section

10.2). Therefore, the fact that a linguistic unit is stored in lexical memory

because it has unpredictable meaning properties, or has a certain frequency

of use, or both, does not tell us anything about its word status.

Thismay be quite obvious, but thematter becomes theoretically interesting

when we consider relatively small units, for instance combinations of two

words that could either be compounds (hence words) or phrases. This prob-

lem shows upquite frequently in the analysis of adjectiveþ noun sequences in

Germanic languages such as those mentioned above: red tape, yellow fever.

Such sequences have the same function as undisputable words: that of

classifying and referring terms. Why do we want to determine whether a

combination of an adjective and a noun is a compound or a phrase? The

answer is that assigning the proper structural interpretation to such word

sequences is essential for predicting their behaviour with respect to modules

of the grammar, such as morphology or phonology. The relevance for the

phonology of English is that in phrases main stress is on the head constituent,

whereas in compounds the non-head bearsmain stress. Thus, we can hear the

diVerence between a dark room (a room for photographic processing) and a

dark room (a room that is dark). The relevance formorphology is obvious for

languages with inXectional marking of adjectives, such as Dutch and

German: the structural analysis determines whether the adjective bears an

inXectional marking. For instance, the translation of the English A þ N

the word as a linguistic unit 283

sequence hard disk is harde schijf inDutch, butFestplatte inGerman. Both the

Dutch and the German expression are AþN sequences. However, in Dutch

the adjective hard is inXected, and ends in -e, whereas inGerman, a language

that also features adjectival inXection, the adjective fest ‘Wxed’ bears no

inXectional marking. This observation leads to the conclusion that harde

schijf is a phrase, whereas Festplatte is a (complex) word, namely an AN

compound. This reasoning is based on the principle of Lexical Integrity.

Remember that this principle claims the following (section 8.1):

(1) ‘The syntax neither manipulates nor has access to the internal form of words’

(Anderson 1992: 84).

The inXectional marking of the adjective in the Dutch example harde schijf

is a case of agreement, a rule that is syntactic in nature. Hence, the fact that

the adjective has the form hard-e (an inXected form of the adjective that

encodes agreement for number, gender, and deWniteness with the head

noun), in combination with the principle of Lexical Integrity forces us to

conclude that harde schijf cannot be a word since otherwise we would have

a case here in which the syntactic rule of agreement manipulates the form of

a word-internal constituent.

This illustrates that we have now achieved some further theoretical

sophistication of the notion ‘word’ by formulating the principle of Lexical

Integrity. This principle is a theoretical interpretation of the deWning

property of the linguistic unit ‘word’ that it cannot be interrupted. Hence,

theoretical notions such as ‘word’ are not isolated concepts that stand on

their own, linked directly to empirical observations. Instead, their content

is determined by their place in the network of terms and hypotheses that

together form a scientiWc theory of a certain empirical domain. Therefore,

we cannot determine purely intuitively, without further theoretical analy-

sis, whether a particular sequence of words is a (complex) word itself: this

requires further theoretical reasoning.

12.2 The demarcation of words and phrases

The issue whether a linguistic unit is a word or a phrase comes up regularly

in relation to compounding: a word sequence might be either a compound

(hence a word) or a phrase. Consider the following examples from French:

284 conclusions

(2) a. pomme de terre

apple of earth ‘‘potato’’

b. robe de chambre

cloak of chamber ‘‘chamber cloak’’

c. clair de lune

light of moon ‘‘moonlight’’

d. machine a ecrire

machine for write.inf ‘‘typewriter’’

e. voiture d’ enfant

car of child ‘‘perambulator’’

The English glosses for these French examples are all (simplex or com-

pound) words. These French expressions are all established expressions for

referring to speciWc entities. In other words, they are listemes of French.

Hence, it is no wonder that people have sometimes called such expressions

compounds. Yet, they clearly have the shape of a French phrase of the type

[N [Preposition N]], and follow the syntactic rules of French that allow for

noun phrases with a nominal head followed by a PP complement. There-

fore, the linguist Benveniste concluded that so-called nominal compound-

ing in French is in fact a kind of what he calls ‘micro-syntax’ (Benveniste

1967). What we observe here is that syntactic patterns that reXect the

syntactic rules of a language function as templates for coining terms.

Note, however, that there are restrictions in this use of syntax for coining

lexical units. The syntactic restriction in French is that, in contrast to

normal syntax, the second noun in this type of lexical unit can normally

not be preceded by an article, or followed by an adjective that only modiWes

the second noun. An expression like voiture d’un enfant is certainly possible

but blocks the speciWc interpretation ‘‘perambulator’’. It has a diVerent

meaning, ‘‘car of a child’’. Similarly, the expression une pomme de terre

noire (lit. ‘‘an apple of earth black’’) denotes a black potato; that is, it is not

the noun terre that is modiWed, but the whole expression pomme de terre.

Similar observations can be made for the A þ N sequences in Germanic

languages. In yellow fever the adjective cannot be modiWed (*very yellow

fever) without the loss of the speciWc disease interpretation, although very

yellow is a correct adjectival phrase of English. Here are some examples of

such lexical multi-word expressions from Spanish and Greek:

(3) Spanish

telon de acero

curtain of iron ‘‘iron curtain’’


luna nueva

moon new ‘‘new moon’’

Greek

psixros polemos ‘‘cold war’’

mavri lista ‘‘black list’’

In many linguistic descriptions of such expressions, they are referred to as

‘syntactic words’, which means that they have the formal status of a syntactic

unit, but are functionally similar to words in the strictly morphological sense.

In traditional grammars of Spanish such expressions are referred to as

‘improper compounds’ which also reXects the hybrid status of these linguistic

units (Rainer and Varela 1992: 120). This shows once more that the notion

‘word’ in the formal linguistic sense has to be clearly distinguished from the

notion ‘term’. A term is a linguistic expression used for the classiWcation and

denotation of entities, but terms are not necessarily words.

The delimitation issue also crops up in the analysis of particle verbs in

Germanic languages such as German and Dutch. The particle verbs in these

languages are similar to those of English. A diVerence with English is that the

particles appear immediately before the verbs in embedded clauses (example

4a) whereas they are stranded in main clauses (example 4b; examples from

Dutch) since Wnite verbs in main clauses appear in second position:

(4) a. dat Jan de informatie opzocht

that John the information up looked

b. Jan zocht de informatie op

John looked the information up

As (4a) illustrates, the rules of Dutch orthography prescribe that particle

and verb are written as one orthographic word, even though they do not

form one word from the linguistic point of view since verb and particle can

be split, as in (4b). Hence, the word in its orthographic sense is not

necessarily a direct reXection of the word in its morpho-syntactic sense.

In conclusion, not everything that native languages users may call a word

is a word from the linguistic point of view.

12.3 Phonological words

In section 7.2.4, the notion ‘phonological word’ has been introduced. The

phonological word is a linguistic unit for the representation of the prosodic

286 conclusions

properties of words and phrases (such as stress pattern and syllabiWcation),

and is also a domain of application for phonological rules (such as vowel

harmony in Hungarian). Recall that the properties of words require three

diVerent forms of representation: a morpho-syntactic, a phonological, and

a semantic representation. The crucial insight for our discussion of the

notion ‘word’ is that we must distinguish between ‘morpho-syntactic word’

and ‘phonological word’. Furthermore, we should realize that these two

types of units are not isomorphic. A Dutch compound, for instance, is one

morpho-syntactic word, but consists of more than one phonological

word. Moreover, some aYxes, that is, bound morphemes from a mor-

pho-syntactic point of view, nevertheless form phonological words of

their own, as shown in section 7.2. The English compound syllable bound-

ary mentioned above consists of two phonological words, syllable and

boundary. Each of these phonological words has a word stress of its own

(with the Wrst one being the most prominent one), and is a domain of

syllabiWcation (the division of words into syllables).

Many English compounds are spelled with internal spaces (referred to as

‘open spelling’), hence as at least two orthographic words. One might say

that in such cases, the orthography reXects the division of that compound

into phonological words rather than its morphological make-up. An

English compound may also be spelled without internal spacing (‘solid

spelling’), or with a hyphen. Hence, we have spelling triplets such as tax

payer, taxpayer, tax-payer. Since there are three options, writers of English

are sometimes uncertain as to what the conventionally correct spelling for a

compound is. The hyphenated spelling, as in tax-payer, indicates both that

the word consists of two units (phonological words), and that it forms one

morpho-syntactic word. Many English preWxes are also separated from the

stem by a hyphen, unlike suYxes: anti-social, co-pilot, mini-skirt, post-war,

pre-eminent, re-integrate. This reXects that English preWxes are phono-

logical words of their own, unlike (most) suYxes. Hyphenation may also

be used to indicate that a syntactic construction has the status of lexical

unit. This is the case for lexical expressions such as do-it-yourself and jack-

of-all-trades.

The diVerence between the morphological and the prosodic structure of

a complex word is also reXected by diVerent orthographic conventions with

respect to hyphenation when a word is distributed across two lines. For

instance, in British English, the word structure is hyphenated as struct-ure,

reXecting the morphological make-up of this word, whereas in American


English this word may be hyphenated as struc-ture which reXects the

syllable division, and hence the prosodic structure of this word.

The English spelling of compounds may be compared to that of Dutch.

In Dutch, a compound has normally to be spelled as one orthographic

word. That is, the spelling reXects the morpho-syntactic representation of

compounds. Many users of Dutch, however, have problems with applying

this rule of orthography properly, and tend to use internal spaces in the

orthographic representation of compounds. That is, they give priority to

the phonological constituency of compounds in the spelling of compounds

(Dutch compounds consist of at least two phonological words).

The theoretical distinction between morpho-syntactic word, phono-

logical word, and orthographic word is thus an important tool for coming

to terms with the behaviour of compounds with respect to morpho-syntax,

phonology, and orthography.

The non-isomorphy between morpho-syntactic word and phonological

word can also be observed for clitics (cf. section 7.2). Clitics (also called

‘small words’) are often qualiWed as being partially word-like, and partially

aYx-like. They are words in that they may occupy the same syntactic

positions in sentences as other words. On the other hand, they form a

prosodic unit with an adjacent word. The best way to deWne the special

status of clitics is that in terms of prosodic deWciency: they are words in the

morpho-syntactic sense, but not in the phonological sense. Whereas,

normally, a word forms a phonological word of its own, a clitic requires

a host word on which it can lean, and with which it can form a prosodic

constituent (it is either part of the phonological word, or adjoined to it).

The Dutch clitic er [

e

r], spelled as ‘r in order to indicate its clitic status, may

serve to illustrate this kind of behaviour. This clitic is the weak, stressless

form of the pronoun haar ‘her’ (and also the weak form of the adverb er

‘there’). This clitic has the same phonological form as the suYx -er that

is comparable to the English suYx -er. The following Dutch sentence

illustrates the use of this clitic pronoun:

(5) Ik vind ‘r lief /Øk vØnd

e

r liv/ [Øk.fØn.t

e

r.lif]

I Wnd her sweet ‘I love her’

The phonetic form of the word sequence vind ‘r is [vØn.t

e

r], with a syllable

boundary between [n] and [t]. The last consonant of the verb that is realized

as a [t] forms one syllable with the segments that form the pronoun ‘r.

That is, the verbal form and the following weak pronoun appear to form

288 conclusions

one phonological word, because the domain of syllabiWcation is the

phonological word.

Dutch also has the suYxed word vind-er ‘Wnder’ derived from the verb

vind ‘to Wnd’. The phonetic form of this word is [vØn.d

e

r], with a [d],

whereas the Wnal consonant of the verb vind in (5) is realized as a [t]. This

diVerence in phonetic realization can be explained as follows: a clitic is

attached to the host word in the syntax, whereas the suYx -er is added to the

verbal stem with underlying Wnal /d/ through a morphological operation,

before the phonetic form of the word is computed. Dutch words are subject

to a phonological rule of Wnal devoicing: obstruents such as /d/ are voiceless

at the end of a syllable. In the word vinder [vØn.d

e

r], the /d/ appears in

syllable-initial position, and hence does not devoice. In vind ‘r, however, the

/d/ is devoiced into [t], before the clitic is attached, in the syntax. That is

why clitics have sometimes been called ‘post-lexical aYxes’. Thus, by

introducing the notion ‘phonological word’ we are able to give a proper

characterization of what a clitic is, and to what extent it is a word. Thus, we

give further substance to the theoretical notion of ‘word’.

12.4 Lexical integrity

The principle of Lexical Integrity given above forbids syntactic rules to

either manipulate or have access to word-internal morphological structure.

The Wrst of these prohibitions (‘no manipulation’) serves to Xesh out the

idea that words have internal cohesion and cannot be interrupted. That is,

it is quite essential for a proper deWnition of the notion ‘word’.

Syntactic agreement is a good testing ground for lexical integrity, as we

saw above. A cross-linguistically well known rule of syntactic agreement is

that of number agreement between quantiWer and head noun. This rule

applies in Spanish: if a quantiWer expresses plural number, the head word of

the phrase with the quantiWer has to appear with plural number as well. In

the case of Spanish copulative compounds, it appears that both parts of the

compound have to be plural. For instance, the plural form of the copulative

compound poeta-pintor ‘‘poet-painter’’ is poetas-pintores (as in dos poetas-

pintores ‘‘two poet(s)-painters’’). This does not imply, however, that a

syntactic rule of agreement of Spanish manipulates parts of words. Instead,

this regularity can be interpreted as a morphological phenomenon: when

we want to pluralize a Spanish copulative compound, both constituents


must be marked for plural. Therefore, this word-internal plural marking is

only an indirect eVect of the syntactic rule of agreement, and hence does

not violate the principle of Lexical Integrity.

Another apparent counterexample to this part of the Lexical Integrity

Principle (prohibition of manipulation of the internal structure of words by

the syntax) is the phenomenon of gapping, illustrated in section 7.2 by

means of examples from Dutch. Some examples from English and Euro-

pean Portuguese (Vigario 2003: 251) respectively are:

(6) a. mono- and polysyllabic

inter- and intranational uses

homo- and heterosexual relations

b. pre- e pos-guerra [¼ pre-guerra e pos-guerra] ‘‘pre- and post-war’’

segura- mas lentamente [¼ seguramente mas lentamente] ‘‘surely but

slowly’’

These are cases of prosodic gapping rather than syntactic gapping: a

prosodic word is deleted under identity with another phonological word

in the same phonological phrase. That is, this process of gapping does not

refer to the morphological structure of words, but to their prosodic con-

stituency. Therefore, this kind of gapping does not violate the principle of

Lexical Integrity.

The second part of the principle of Lexical Integrity refers to accessibil-

ity. It excludes that syntactic rules make use of information about the

internal morphological composition of a word. Let us have a look at

some phenomena that suggest that this part of the principle of Lexical

Integrity is too strong.

One of these phenomena is what we might call ‘context-dependent

morphology’: the situation in which a particular syntactic construction

goes hand in hand with a particular morphological form of words in that

construction. The case that I will discuss here to illustrate this point is the

pluralization of Dutch numerals. Dutch features plural forms of ordinal

numerals and some other quantiWers. The use of the plural forms of most

numerals is restricted to a number of speciWc constructions, which are

exempliWed in (7):

(7) a. number of parts

Het schip brak in drie-en

The ship broke into three-en

‘‘The ship broke into three pieces’’

290 conclusions

b. appositive collective

wij / ons drie-en

we / us three-en

‘‘the three of us (subj. / obj.)’’

c. collective adverbial

met ons / jullie / hun drie-en

with us / you / their three-en

‘‘the three of us /you / them together’’

d. collective adverbial

met z’n [z

e

n] drie-en

with his three-en

‘‘the three of us / you / them’’

Example (7d) is a prototypical case of a constructional idiom (cf. section

8.4). It has the form of a PP, headed by the preposition met, followed by

the NP [z’n Numeral-en]. In this NP the slot for the possessive pronoun is

Wxed as z’n (the clitic form of the 3sg possessive pronoun), whereas the slot

for the numeral is open and can be Wlled with all sorts of numerals, as

illustrated in (8):

(8) We komen morgen met z’n twintig-en

We come tomorrow with his twenty-en

‘‘We will come tomorrow with twenty persons’’

Note the incongruence between the person and number of the subject (1pl)

and that of the possessive pronoun (3sg). The examples in (7c) are variants

to be used with agreement in person and number between the subject of a

sentence and the possessive pronoun in the collective construction. So there

are two diVerent collective constructions that are identical except that the

possessive pronoun can either be a variable (and thus subject to the normal

agreement constraints for possessive pronouns), or a Wxed possessive

pronoun z’n ‘his’.

You may have noticed that in the glosses for the Dutch examples I did

not make use of the morpho-syntactic feature plural, but mentioned the

concrete Dutch suYx -en instead. The reason for this is that it is indeed the

speciWc plural suYx -en that is required by this construction. Dutch has two

competing plural suYxes, -s and -en. The choice between these two suYxes

is determined by a prosodic output condition: the plural noun should end in

a trochee. Hence, the suYx -en is selected after stems ending in a stressed

syllable, and the suYx -s appears after stems ending in an unstressed


syllable. However, in the numeral constructions discussed here, the suYx is

always -en, even when the last syllable is unstressed. For instance, the plural

for the number 7 is zeven-s /zev

e

ns/, as is the case when we talk about

grading (Jan kreeg twee zeven-s ‘‘John got two grades 7’’). Yet, in the uses

shown in (7), the plural of zeven is zeven-en /zev

e

n

e

n/. The same applies to

the number negen ‘9’ /neg e

n/: its plural is normally negen-s, but in these

constructions it is negen-en.

These observations imply that we have to specify the presence of a

speciWc suYx -en in the constructions exempliWed in (8). For instance, the

constructional pattern for phrases such as that in (7d) is:

(9) [[met] P [z ’n [ [x]Numeral -en]N]NP]PP

with his x-pl

‘‘the x of us / you / them’’

This analysis suggests that the principle of Lexical Integrity as formulated

in (1) is indeed too strong, and that the syntax may require access to the

internal morphological structure of words.

Another example that implies that such a weakening of the principle of

Lexical Integrity is necessary comes from Georgian. In Georgian we Wnd

expressions like the following (Harris 2006):

(10) sam tit-moc’r-il-i (k’ac-i)

three.obl Wnger-cut.oV- ptcp-nom man-nom

‘‘(a man) with three Wngers cut oV ’’

The Wrst word sam ‘‘three’’ has to appear in the oblique form, because it

modiWes the word tit ‘‘Wnger’’ which is part of the secondword. That is, both

for the purpose of case assignment and semantic interpretation, sam and

tit form a unit. Hence, case assignment requires access to the internal

morphological structure of the second word in (10). The construction in

(10) may be compared to that in (11) where the Wrst word bears nominative

case, and where one gets a diVerent interpretation:

(11) sam-i tit-moc’r-il-i

three-nom Wnger-cut.oV- ptcp-nom

‘‘three (men, people, statues) with Wngers cut oV ’’

In (11), the word sami agrees in case marking with the second word as a

whole, and hence it is a modiWer of the whole word.

292 conclusions

As far as semantics is concerned, the Georgian example is similar to

the English phrase transformational grammarian in which the adjective

transformational modiWes the constituent grammar of the complex word

grammarian (cf. section 9.1 where another example, moral philosopher, is

mentioned). The additional property of the Georgian construction is that

case assignment is also sensitive to word-internal structure.

These phenomena thus imply that the principle of Lexical Integrity as

stated in (1) should be weakened. However, this is no blow to our notion

‘word’ as a cohesive and non-interruptable unit of linguistic structure.

Summary

In this chapter we have seen how the theoretical interpretation of the

notion ‘word’ has been developed in this book. A theoretical term is

embedded in a network of theoretical distinctions and hypotheses. The

ingredients that we used for developing an adequate characterization of the

notion ‘word’ are the distinction between lexeme and word-form, and that

between words and lexical unit/listeme, a proper demarcation of syntax and

morphology, and the distinction between morpho-syntactic, phonological,

and orthographic words.

The principle of Lexical Integrity as formulated in (1) can be seen as

a theoretical interpretation of the notion ‘wordhood’. It is too strong,

however, because syntax (and also semantics) may require access to the

internal structure of complex words. The observations and analyses in this

chapter show once more that the grammar of words and the grammar of

sentences are intertwined in many ways!

Further reading

The importance of distinguishing the different notions ‘word’ discussed in thischapter is also argued for in Dixon and Aikhenvald (eds.) (2002, in particular inthe first introductory chapter of that volume). The demarcation of wordsand word-like phrases in a range of languages is discussed in Chapter 10,‘Incorporating patterns’ of Dahl (2004). The importance of the notion ‘phono-logical word’ is demonstrated in Hall and Kleinhenz (eds.) (1999). Nevis (2000)provides a survey of clitic phenomena, and a recent comprehensive theoretical


study of clitics is Anderson (2005). Amore detailed analysis of the phonologicalbehaviour of Dutch clitics is given in Booij (1996b). Particle verbs in Germaniclanguages are discussed in van Kemenade and Los (2003). Context-dependentmorphology is discussed in Booij (2005). The Georgian data and theirimplication for the principle of Lexical Integrity are discussed in Harris (2006).

294 conclusions

Answers to questions

Chapter 1. Morphology: basic notions

1. Supposedly has the not completely predictable meaning ‘‘most likely,presumably’’.

2. The pronunciation and meaning of -able as an affix are different fromthose of able as a word.

3. [[un[happi]A]Aness]N, [[contrast]Nive]A, [[dis[connect]V]Ving]V,[[contradict]Vion]N, [[blue]A[[eye]Ned]A]A or [[[blue]A[eye]N]Ned]A,[[[connect]Vive]Aity]N

4. col-league or colleague, cord-ial, cor-rel-ate or cor-relate, electr-o-meter, e-long-at-ion, e-vapor-ate, etern-ity, eu-phem-ism or euphem-ism, habit-ual orhabitu-al, happ-y or happy, mus-ic or music, negoti-able, per-form-ance orperform-ance, theo-logy or theo-log-y.

5. Plural endings: -oollee, -oota, and -eellee.

6. (i) There are corresponding simplex verbs bid, get, give, go, and swear. (ii)There is a recurrent element for. (iii) They have the same strong inflectionas the corresponding simplex verbs.

7. All these words indicate that something comes out of something.

8. This is a case of analogical word-formation: mother: father ¼ mothertongue: father tongue.

9. Falloween: an extended celebration or observance of Halloween, often be-ginning several weeks before the day; the retail season that extends from thebeginning of fall through Halloween and Thanksgiving in the US (fall þHalloween). Giraffiti: graffiti painted at a high spot (graffiti þ giraffe). Metro-sexual: an urbanmale with a strong aesthetic sense who spends a great dealof time and money on his appearance and lifestyle (metropole þ heterosex-ual ). Nicotini: a nicotine-laced martini (nicotine þ martini ). Pedlock: thecondition or state of being so crowded that people are unable tomove easilyin any direction ( pedestrian þ gridlock).

10. a blending (back þ acronym).

Chapter 2. Morphological analysis

1. disagreeable: dis-, -ableacceptability: ac-, cept, -able, -ity

ungrammaticality: un-, gram, -at, -ic, -al, -itydiscriminatory: dis-, crimin, -at, -orypermafrost: perma-fascination: fascin, -ationprotolanguage: proto-versification: -ific-, -at, -ionintolerance: in-, toler, -anceunidirectionality: uni-, -ion, -al, -ity

2. a. The mid front vowels [œ] and [E] become back vowels [] and [a]respectively before the suffixes -al, -itude, -aire, -ite, -ifier, -in, and -iste.

b. This is a morphonological rule because its application is conditionedby the presence of specific suffixes.

3. a. [t], [d], [Ød].

b. Assume an underlying /d/ as the phonological form of the past tensesuffix. The /d/ is devoiced and thus becomes a [t] if the stem ends in avoiceless consonant; the vowel [Ø] is inserted before the /d/ if the stemends in /d/ or /t/.

4. ki-tapin-a:-wa:w.

2PERS-sit-PL–2PERS‘‘You (plural) sit’’

5. Suffixation and vowel change.

6. a. a a a a a a| || | | |

CVCCVCV CVCCVCV| \/ | | | | |b k l t l f n

b. a a a| | |

CVCCVCV

7. Plural suffix is -i. The last vowel of the stem is deleted in plural nouns with astem ending in /r, n, m/. Stem-final /k/ is voiced before the plural suffix.

8. Past Active: infixation with -um-; Past Passive: infixation with -in-; PresentActive: Partial Reduplication and infixation; Present Passive: Partial redupli-cation and infixation.

9. Copy the first (C)VC of the stem, and prefix this copy.

10. fam, giall, virtu, blu.

296 answers to questions

Chapter 3. Derivation

1. The verb to joyride is a conversion of the noun joyride, with the structure[[[ joy]N[[ride]V]N]N]V. Hence, the feature [þablaut] of the verbal root ride,from which the head noun ride of the nominal compound joyride is derived,cannot be percolated to the highest V node, because there are interveningN nodes that cannot carry such a feature.

2. The suffix -able can be attached to verbs of Germanic origin such as do andread: doable, readable.

3. The prefix en- appears to turn nouns and adjectives into verbs. Thus, itseems to determine the category of the prefixed word, and to function ashead, although it is in left position. This is a problem for the Right-handHead Rule, unless one assumes that the base nouns and adjectives have firstbeen converted to verbs, and have subsequently been prefixed with en-.

4. One may assume that the prefix in- derives verbs from adjectives, and thatsuch verbs are automatically assigned to the default conjugation, just likethe converted nouns dribblare and scioccare. Hence, the appearance of thethematic vowel -a- is a predictable effect of the assignment to this conju-gation, and not due to suffixation with -a as part of a parasynthetic word-formation process.

5. In these nouns there is no base noun with a gender specification that can bepercolated to the dominating node.

6. This use of reduplication is not iconic, since we would then expect anintensified meaning for such reduplicated adjectives.

7. The prefix de- has a privative or a reversative meaning, and attaches to verbsand nouns. The prefix dis- mainly attaches to verbs, with reversative orprivative meaning, and may also express negation. Of these two, the prefixdis- is the preferred one before vowel-initial stems. Both combine withnon-native base words only. The prefix in- has a negative meaning andattaches to non-native adjectives only. The prefix non- attaches to adjectivesand nouns, with negative meaning. It differs from un- in that it preserves therelational character of its adjectival base. The prefix un-, with the meaning‘not’ can be attached to adjectives, verbs, and nouns, of both native andnon-native origin.

8. a. No, we might consider this suffix an adjectival suffix, that assigns thecategory A to the diminutive adjective.

b. This suffix cannot be attached to relational adjectives since it has aqualifying meaning.

9. The prefix a- appears before consonant-initial stems, the prefix an- beforevowel-initial stems. Thus vowel hiatus (two vowels in adjacent position) isavoided.

answers to questions 297

10. This is a case of type coercion: American has to be interpreted as aqualifying adjective.

Chapter 4. Compounding

1. NN computer deskAN blackboardVN pickpocketPN undergroundAA dark-blueNA ice-coldVV freeze-dryAV blindfoldNV brainwash

The heads of English compounds are N, A, or V. Prepositions only combinewith N heads. Vs do not combine with A heads.

2. [[[re[create]V]Vion]N[hall]N]N[[book]N[[keep]Ving]N]N[[truck]N[[driv]Ver]N]N[[pick]V[pocket]N]N[[under]P[dog]N]N[[home]N[[grown]V]A]A

3. The left constituents function as heads of these compounds.

4. Although gate is a lexeme, it has a completely different meaning whenused as part of these complex words. Hence, we might consider thesewords as cases of derivation with the affix -gate.

5. In (5b) the noun is incorporated. This results in a verb that denotes aninstitutionalized act, tree-chopping. The incorporated noun is non-refer-ential, hence the object in-kool ‘‘my cornfield’’ is possible.

6. The first A þ N combination is a compound, the second one a phrase.

7. In this compound neither of the constituents functions as the head, andtherefore it is exocentric.

8. bear jam: a traffic jam in a park caused by motorists stopping to watch oneor more bears;

deprivation cuisine: food prepared in such a way that it is healthy, butbland;

flash mob: a large group of people who gather in a usually predeterminedlocation, perform some brief action, and then quickly disperse;

information pollution: the contamination of a culture or of a person’s lifecaused by exposure to excessive amounts of information or data;


man breasts: excess fatty tissue that causes a man’s chest to resemble awoman’s breasts;

office creeper: thief who walks into the workplace looking for pricey laptopsand purses;

salad dodger: an overweight person, a person who shuns healthy foods.The following words are hard to interpret on the basis of the meanings ofthe constituent words: bear jam, deprivation cuisine, flash mob, officecreeper, salad dodger.

9. Such coordinative compounds do not have a head.

10. This is a case of coordinative compounding by means of full reduplication.In addition, the first consonant of the second constituent is replaced withsh; this segment is added before a vowel-initial second stem.

Chapter 5. Inflection

1. This is an elliptical construction with the accusative case assigning verbomitted. It means ‘‘I wish you a good morning’’.

2. The synthetic form is used for monosyllabic adjectives (big–bigger), andbisyllabic adjectives with a final light syllable (happy–happier). In othercases the periphrastic form is used, as in more excellent/*excellenter andmore meaningful/*meaningfuller.

3. This depends on your theory of inflection. In a realizational theory ofmorphology you do not have to, because there will be no rule that istriggered by the presence of the feature [infinitive].

4. These facts show that inflected nouns can form constituents of com-pounds, and hence feed word-formation.

5. The two suffixes in anno-i-n are cases of inherent inflection, whereas theother endings are cases of contextual inflection. The case markings on vet-ta and koira-lle are due to structural case assignment (government), thecase marking on kahde-lle is an instance of agreement.

6. This is quite tricky because it means that suppletion will be applicable to allpairs of different lexemes with similar, related meanings (for instanceman–woman, bicycle–moped), and hence the notion suppletion will losemuch of its distinctive power.

7. The GEN.SG form is a case of cumulative exponence since the suffix -eexpresses both GENITIVE and SINGULAR. This form is a case of extendedexponence as well, since the vowel alternation and the alternation in thestem-final consonant also mark GEN.SG. The DAT.SG form is also a case ofextended exponence, since the property cluster DAT.SG is expressed byboth a vowel alternation and a consonant alternation.


8. In the British English lexicon, the word police must be specified as[þplural], in the American English lexicon as [�plural].

9. MASCULINE þ FEMININE ¼ MASCULINE.

10. Panini’s Principle. The rule that suffixes -eren will be subject to a morespecific condition since it has to refer to a diacritic feature, say [þeren], andhence takes precedence over the more general rule of -en-suffixation.

Chapter 6. Inflectional systems

1. The cut-off point for the threefold distinction is between animate andinanimate. Human beings and (larger) animals can be referred to withhe and she, but for inanimate entities, only it can be used.

2. a. Stem shapes for kasi: kas (nominative, comitative), kade (genitive,translative, abessive, inessive, elative, adessive, ablative, allative, instru-mental), katt (partitive), kate (essive, illative); for tytto: tytto (nominative,partitive, essive, illative, comitative), tyto (all other cases).

b. tyton.

3. a. vyr-, broli-, arkli-

b. NOM.PL ¼ VOC.PL.

4. The notions periphrasis and suppletion. In the periphrastic forms bothstems are different from that of the synthetic forms.

5. Past tense is used here for creating distance, hence indirectness, and thuspoliteness.

6. In plural forms, NOM ¼ ACC ¼ VOC.

7. The verbal infinitive can function as a neuter noun; hence, it will triggergender, person, and number agreement between the subject and thepredicate.

8. The numeral functions here as the head of the phrase since it imposesgenitive or ablative case on the constituent cocuk-lar ‘‘children’’. Hence,the genitive or ablative case marking on cocuk-lar must be a case ofdependent marking.

9. The presence of the time adverbial morgen ‘‘tomorrow’’ in the Dutchsentence already indicates that the event will take place in the future.

10. Coughing is a punctual event, and is only semantically compatible withprogressive aspect if the event is repeated.


Chapter 7. The interface between morphology

and phonology

1. This can be concluded from the syllabification of words with these suffixes,for instance: rea.da.ble, wor.ker, wor.king. Crucially, the word-internal syl-lable boundaries do not coincide with a morphological boundary.

2. These function words do not contain a full vowel, and hence no syllable thatcan bear stress. Thus, they cannot form a foot, and hence no prosodic wordof their own, which is a requirement for lexical words. Therefore we canconclude that they must be function words.

3. The /n/ also occurs in other derived words from the same root, such asPlatonist and Platonism. By considering the /n/ as part of the stem, wepredict this systematic appearance before (non-native) suffixes.

4. We might explain this by assuming that the domain of this rule of assimi-lation is the phonological word. The prefix in- can be considered as acohering prefix, and the prefix non- as a non-cohering prefix. Hence, therule of assimilation will only apply to the prefix in- since it forms onephonological word with the stem, a domain in which the assimilation rulecan apply.

5. This can be shown by making a tableau, for instance for the word bak-erwith the morphological stem bake:

[[bak] er] ALIGNMENT-LEFT NO EMPTY ONSET ALIGNMENT RIGHT

ba.ker

bak.er

F

*!

*

6. In this case, the compound constituent beren is deleted under identity withan independent noun beren. Therefore, these constituents are syntacticallynot identical, and cannot be gapped if this process is conditioned bysyntactic identity. Hence, their identity must be identity on the prosodiclevel where they both form a prosodic word (beren).

7. This change follows from the requirement that a lexical word is minimally aprosodic word. Hence, each lexical word requires the presence of at leastone full vowel. Thus, the letter i that stands for schwa in the suffix isreinterpreted as a full vowel [Ø].

8. This rule will give the right result for breakfast table because both constitu-ents are disyllabic. As soon as a compound has longer constituents, wrongpredictions are made. In hıstory teacher, with secondary stress on the syl-lable tea, this rule would assign stress to the last syllable of history instead ofthe first syllable of teacher. Therefore, the two parts history and teachermustform independent domains of stress assignment.


9. This suffix, or the vowel /e/ of this suffix, must be qualified as transparent:it does not affect vowel harmony, otherwise the dative suffix would be -nekin both words. The choice of the vowel is clearly determined by the vowelsof the roots of these words.

10. The determiner a is selected before a consonant-initial word, the deter-miner an before a vowel-initial word. Since the English determiners formone phonological word with the following noun, the choice of an, as in anarm, with the syllabification a.narm, avoids violation of the No EmptyOnset Constraint that would be violated in a arm with the syllabificationa.arm. The choice of a before a consonant-initial word avoids a violation ofthe Open Syllable constraint: in an book with the syllabification an.bookthe first syllable is closed, unlike the first syllable in a book with thesyllabification a.book.

Chapter 8. Morphology and syntax: demarcation

and interaction

1. The intended referent of it in sentence (a) is truck. This latter word is herepart of a compound, and thus the word truck is not directly accessible forrules of anaphora because of Lexical Integrity. In sentence (b), the pro-noun one refers to the whole word truck driver, and hence we do not haveto violate Lexical Integrity to establish the relevant anaphoric relation.

2. The phrase small claims embedded in a compound has a classificatoryfunction: a small claims court is a court for a particular kind of claims,small claims. The phrase small claims has therefore a classificatory function.Hence the adjective small cannot be modified by very.

3. No, lexical units larger than one word can also feed word-formation.

4. The Causer argument of the causative verb is the highest argument, andwill be assigned the grammatical function of subject (in imperativeclauses, the subject is usually not expressed, but understood as the ad-dressee). Hence, the Agent argument will receive the grammatical func-tion of object, and therefore be marked with accusative case.

5. a. The verb for ‘‘to begin’’ must be lexically specified as imposing dativecase on its object, since the default case marking for objects is accusative.

b. In the passive sentence, the noun dersler ‘lessons’ is not case markedas a Subject, but keeps its dative case marking. Hence, it is not a Subject,unlike what we would expect if passivization were the promotion ofPatient to the grammatical function of Subject.

6. This is a case of valency increase.

7. The addition of the assistive derivational morpheme creates a verb withthree arguments, and hence there are two arguments that are non-sub-


jects. Both ‘‘mother’’ and ‘‘brother’’ are marked with the appropriate casefor non-subjects, the accusative case.

8. This is a case of applicative verb formation in which the beneficiary of theaction is also expressed. Hence, it is an instance of valency increase.

9. The combinations of these verbs with past participles can be considered asthe periphrastic fillers of passive cells of the verbal paradigm. The specific,non-compositional meaning of these word combinations will be assignedto these cells. You may also consider these combinations of the verbsworden and zijnwith a past participle as constructional idioms. The passivemeaning and the unexpected perfective interpretation of the imperfectiveforms of zijn þ past participle will then be considered properties of theseconstructional idioms.

10. The two verbs in this sentence, liep and dede cont, apparently function as averbal unit. We do not have two coordinated clauses here, but one clausewith a coordinated verb cluster. The object dit ‘‘this’’ of dede cont thusbecomes an object of this multi-word unit, and therefore precedes this unitwhen it is fronted.

Chapter 9. Morphology and semantics

1. Doctor who treats horses; doctor (gardener) who treats trees; doctor whomakes use of voodoo rituals; doctor employed in a hospital.

2. beehive: metaphor (place where many people come together)bottleneck: metaphorconvention: metonymy (activity > set of people involved in the activity)cliffhanger: metaphor (end of television soap with situation of suspense)holder: metaphor or metonymy (person > instrument or location)opposition: metonymy (activity > people involved in the activity)printer: metaphor (person > impersonal agent > instrument)

3. rural policeman: the location of the policeMartian expedition: the goal of the expeditionearly riser: the time of risingindividual decision: the agent of deciding

4. The correspondence rule cannot be universal because in Malagasy thesemantic interpretation of completely reduplicated forms is not ‘‘in-creased’’, but ‘‘decreased’’.

5. The use of selection as a predicate of John implies a result noun interpret-ation. Hence it cannot be combined with an agent phrase.

6. The deverbal noun inherits the case-assigning properties of its base verb.

7. This is no violation of the Non-Redundancy Constraint since theconstituent parkeer makes the interpretation of the Dutch word garage


unambiguous: as part of this compound, it cannot be interpreted as aservice station.

8. The scope of this suffix which denotes ‘‘possibility of undergoing anaction’’ is: x divide by three. It means: ‘‘for this number x, it is possible todivide x by three’’. Hence the scope is larger than just the verb divide.

9. blessing: action and result of actioncrossing: action and location of actiondrawing: action and result of actiondwelling: action and location of actionopening: action and result of action

10. This follows from the polysemy of diminutive suffixes. The first occurrenceof a diminutive suffix in these words denotes the small size, the secondoccurrence signals endearment. Hence, these diminutive suffixes havedifferent meanings, and can therefore be stacked up.

Chapter 10. Morphology and psycholinguistics

1. Dictionaries do not always specify the phonological form of a word (onlythe orthographical form). The mental lexicon will also contain individualassociations with words based on individual, personal experience. Forinstance, in your personal lexical entry for dog you may have your ownprototype of a dog, or an association with fear.

2. These children do not yet command the established, conventionalizedwords for these concepts.

3. The words with a relatively low base frequency: uncanny, unleash, unob-trusive, and uncouth. This correlates with a relative low degree of semantictransparency of these words.

4. These verbs obey the principles of simplicity (conversion is a simple formaloperation) and transparency (meaning can be grasped easily). The prin-ciple of conventionality will have no impact due to the small size of themental lexicon of children.

5. The verb to joyride is a conversion of the compound noun joyride, andhence has no irregular verb to ride as its head. Hence, the irregular pasttense form rode is of no relevance here, and the regular rule of past tenseformation will apply. If such a rule did not exist, we would expect theassociation between ride and rode to result in the past tense form joyrode.

6. Words that have identical word-initial phonological strings are associatedin the mental lexicon. Hence, they might be exchanged.

7. This correlates with token frequency. One refers more often to more thanone tree, hair, tooth, foot, and sister than to a singular tree, a singular hair,etc. Thus, the plural forms of the corresponding Italian nouns have a


higher token frequency, and hence a shorter reaction time than thesingular forms. In the case of ‘‘nose’’, speakers will more often refer to asingular nose than to noses. For soldato and piazza the pragmatic differ-ences between the singular and the plural notions are minimal. Hence wedo not find large differences in response latency for their singular andplural forms.

8. The basis for these errors must be the semantic relationship of (near)synonymy.

9. The past tense forms of the verbs involved differ in that it is made by vowelchange in the case of to know, and by means of suffixation in the case of tohear. It is the abstract feature [past] that must have been transferred fromto hear to to know, not the suffix -d. Otherwise we would have expectedthe form knowed in the if-clause.

10. This shows that patterns that hold for a more or less closed set of wordsonly, are nevertheless recognizable as such for the language user.

Chapter 11. Morphology and language change

1. The non-native suffix -ity can be added to an adjective in -able, whereas -itycannot be attached to native adjectives such as yellow (*yellowity).

2. If plural forms are used more frequently than singular forms, they will havea higher activation level than singular nouns, and paradigmatic levellingwill also go in the direction of plural forms.

3. This is in accordance with the tendency in natural languages that adifference in form implies a difference in meaning.

4. The word shepherd has lost its semantic transparency, and has become asimplex word. Hence, its phonological form has developed into that of asimplex word (a phonological word consisting of one trochee). The sec-ond syllable does not bear stress anymore, and its vowel has been reducedto schwa.

5. If this word were still phrasal, we would expect the plural form pomi d’oro.Thus, the plural form shows that this expression has become a simplexword.

6. Synonyms are violations of this principle.

7. The accusative form requiem is the first word of the opening line of theMass for the Dead:

Requie-m aetern-am dona eis Domin-eRest.FEM-ACC.SG eternal-FEM.ACC.SG. give-IMP.SG them Lord-VOC

‘‘Give them eternal rest, o Lord’’

This first word became the name of the whole Mass.


8. The first constituent of this word is a verbal stem, and hence it cannot havea genitive ending since verbs do not have case.

9. In most contexts, information about number of nouns is non-redundant.Gender has no clear semantic function, and can therefore be dispensedwith easily. Case marking has a semantic function but can be dispen-sed with as well, because word order and prepositions may take over therole of marking grammatical functions.

10. This change is the same as that in which the English genitive suffix -s wasinvolved: it became a clitic (phrasal affix). This is therefore a case ofdegrammaticalization.


Glossary of terms

abessive case case that expresses ‘absence of, distance from’

ablative case case that expresses ‘movement from’

ablaut pattern of vowel alternation in the roots of inXectionally or derivationally

related word(form)s; a case of internal modiWcation

absolute-ergative system system of case marking in which the subject of intransi-

tive verbs and the object of transitive verbs receive the same absolutive case

marking; the subject of transitive verbs has an ergative case marking

absolutive case case marking of the single argument of a verb with one argument,

and of the Object argument for verbs with more than one argument in an

absolutive-ergative system

accusative case case that marks direct objects of verbs

acronym combinations of initial letters of a word sequence that is pronounced as

a word, not as a combination of letters

action nominalization nominalized form of a verb that expresses an action

action noun noun that expresses an action

activation level the degree of activation of a word in the mental lexicon

activation spreading the spreading of activation from one word to another related

word in the mental lexicon

active voice pattern of linking between arguments of verbs and semantic roles in

which the function of Agent is linked to the grammatical function of subject,

and that of Patient to the grammatical function of object

adessive case case that expresses ‘position at’

adjectivalization the formation of adjectives through a morphological process

adjunct a participant in an event that is expressed optionally

adverbialization the formation of adverbs through a morphological process

aYx a bound morpheme, a constituent of a word attached to a stem of a

particular word class that cannot appear as a word of its own

aYx substitution morphological process in which words are formed by replacing

one aYx of the base word with another one

aYx telescoping thehistorical process inwhich twoadjacentaYxesbecomeoneaYx

aYxoid word that exhibits an aYx-like behaviour

agent semantic role of the entity that performs an action

agglutinative language language with complex words in which each morpheme

corresponds with one semantic unit

agrammatism the inability of aphasic patients to use linguistic rules

agreement correspondence in morpho-syntactic properties between words of a

sentence in a speciWc structural conWguration

Aktionsart the lexical or inherent aspect expressed by a verb

alignment the alignment between diVerent structural representations of a word

such as its prosodic and its morphological representation

allative case case that expresses ‘movement towards’

alliterative concord agreement inwhich the same aYx is used for the agreeingwords

allomorph one of the phonetic forms of a morpheme

allomorphy the phenomenon that a morpheme has more than one phonetic

shape; thus, two or more morphs may correspond with one morpheme

alphabetism combination of the Wrst letters of words pronounced with the phon-

etic values of these letters in the alphabet

analogy the pattern of correspondence between similar words or constructions

that enables the language user to make new words or constructions in con-

formity with that pattern

analytic form a word sequence that has the same function as a morphologically

complex word

anaphora the phenomenon that two expressions in a sentence or text refer to the

same entity or action. Used in particular in cases where the referent of a

pronoun is supplied by a preceding referring expression

Animacy Hierarchy scale for the degree of animacy of nouns that plays a role in

the morphological marking of properties

anti-passive morphological marking of verbs in constructions in which the se-

mantic role of Object is not expressed as an argument

apophony pattern of vowel alternation in the roots of inXectionally or deriva-

tionally related word(form)s; a case of internal modiWcation

applicative morphological marking of verbs in constructions in which a partici-

pant that is not expressed as argument in the corresponding base verbs is

promoted to the status of Object-argument

appositive compound compound with a relation of apposition between its con-

stituents

argument noun phrase that receives a semantic role from a verb

aspect the way in which situations (states or events) are presented as to their

internal temporal constituency

atelic aspect aspect that expresses lack of an endpoint of the action or event

attenuative form morphological form that is used to express notions of vagueness

such as ‘more or less, sort of ’

autosegmental morphology the expression of morphological categories through

independent tiers of phonological representation such as the CV-tier or tone

sequences

308 glossary

back formation word formation process in which a less complex word is derived

from a more complex word

bahuvrihi-compound compound that denotes the person in possession of the

entity denoted by the compound

base-driven restriction restriction on a word formation process imposed by

the base words for that process

base frequency the frequency of the base of a word form or derived word with

that base

base word the word to which a morphological operation applies

bimoraic syllable a syllable with a rhyme that consist of two morae

binding the situation in which the interpretation of a variable is Wxed by some

linguistic element

binyan set of forms in the paradigm of Semitic verbs with the same abstract

consonant-and-vowel pattern

bleaching semantic change in which the meaning of a grammaticalized element

becomes more abstract

blend combination of the Wrst part of a word with the second part of another

word

blocking the blocking of the formation of a word through some morphological

process due to the existence of another linguistic unit with the same meaning

borrowing the phenomenon that languages take words from the lexical stock of

other languages as an eVect of language contact

bound morpheme morpheme that only occurs in combination with other mor-

phemes within a word

bracketing paradox the phenomenon that two non-isomorphic structures are

necessary for a complex linguistic expression in order to do justice to its

properties at diVerent levels of representation

breaking the alternation between long vowels and diphthongs in sets of morpho-

logically related words in Germanic languages

case the marking of nouns that encodes their relationship to other elements in

the sentence, in particular verbs, prepositions, and other nouns

category-changing process a morphological process is category-changing if the

syntactic category of the output words diVers from that of the base words

category-determining process morphological process that determines the

syntactic category of its output words

category-neutral process morphological process that preserves the syntactic

category of its input words

causative type of verb with as subject argument an agent that causes an event to

happen

glossary 309

circumWx aYx that is a combination of a preWx and a suYx

citation form the form of a word in which it is cited, for instance in dictionaries

class-changing process a morphological process that changes the syntactic cat-

egory of its input words

clipping process in which words are made by removing a part of that word and

using the remnant as a word with the same meaning, as in mike from micro-

phone

clitic word-like linguistic element that is dependent on a host word to which it

can be attached prosodically. A clitic may have a restricted or special syntactic

distribution compared to other words of the same grammatical category

coda the part of a syllable after the nucleus

cohering suYx suYx that forms a prosodic word with the (last part of the) stem

to which it attaches

combining form a morpheme that is not an independent lexeme, and appears in

neo-classical compounds or is attached to another word; in the latter case they

are similar to aYxes

comitative case case that expresses ‘accompaniment’

common gender non-neuter gender

comparative form of an adjective that expresses comparison, in particular the

meaning ‘more than’

competing aYx aYx that competes for its distribution with a synonymous aYx

complex word word with an internal morphological structure

composition a synonym of ‘compounding’

compositional function the semantic function for a linguistic construction that

computes the meaning of that expression on the basis of the meanings of its

constituents

compound a word that is itself the combination of two or more words

compounding the word formation process in which compounds are formed

concatenation the combination of elements into a linear sequence

concatenative morphology morphology that makes use of the operation of

concatenation

conceptualization rule rule of semantic interpretation that further develops the

interpretation of a complex linguistic expression

concord a synonym of ‘agreement’

conXation the combination of two word formation templates into a new, more

complex one

conjugation a class of verbs that show the same pattern of inXection

connectionism a psychological theory in which knowledge is modelled in terms of

a neural network with connections between pieces of information

constructional idiom an idiomatic pattern in which one or more slots is variable

310 glossary

construct state form of a noun that shows its relation to another element in the

same construction

contextual inXection inXection that is required by the syntactic context in which a

word occurs

converb a verbal form used as an adverb

conversion the making of new words by using words of one syntactic category as

words of another one, without overt morphological marking

co-phonology phonological systemthatapplies toasubsetof thewordsofa language

copulative compound compound with a relation of coordination between its

constituents

coreferentiality the phenomenon that two linguistic expressions are used to refer

to the same entity in the relevant domain of discourse

corpus a set of linguistic data

correspondence rule rule that speciWes the relation between the form and the

meaning of a word

cranberry morpheme non-aYxal morpheme that only occurs as part of some

established complex words, as cran in cranberry

creole language a language originating from a pidgin language that has become

the native language of new generations of speakers

cumulative exponence the simultaneous marking of more than one grammatical

category through one morphological element

cumulative frequency the summed frequency of a set of morphologically related

words or word forms

cyclicity the form of rule application in which a set of rules applies cyclically

dative case case that marks the indirect object of a sentence

declension a class of nouns or adjectives with the same inXectional pattern

deXection the loss of inXectional marking of words

degrammaticalization the inverse of grammaticalization

degree the morphological marking on adjectives of diVerent degrees of presence

of a property

deixis the reference of elements as determined by pragmatic factors such as time,

space, speaker, and addressee

demotion the eVect of a morphological operation in which the participants of

events expressed by verbs are demoted to a lower argument status, or to

adjunct status

dental preterite preterite form with a suYx that contains a dental consonant

dependentmarking morphologicalmarkingof thedependentword ina construction

dephonologization process of language change in which a phonological alterna-

tion loses its original phonological motivation

glossary 311

deponent verb verb with an active meaning but a passive morphological form

derivation the formation of complex words by means of aYxation or non-

concatenative morphology

desinence inXectional ending

desyntactization the loss of syntactic status of a linguistic construction

deverbalization the loss of verbal properties of a nominalized verb

diacritic feature non-predictable feature encoded on a lexical item to indicate its

behaviour with respect to a linguistic rule

dictionary reference book with information on the lexical stock of a language

diminutive word with a morphological marking that encodes relative smallness of

size or high degree of aVection

direct case a synonym of ‘structural case’

domain shift the use of a word belonging to a semantic domain in another related

semantic domain

drift long-term change of a language in a certain direction

dual the morphological marking for the number ‘two’

dual system model of morphological knowledge in which perception and

production of complex words make use of two routes, a direct route with

retrieval of these complex words from lexical memory, and an indirect route

in which these complex words are parsed into their constituents (perception) or

put together on the basis of rules (production)

durative aspect aspect indicating that an event has a certain duration

dvandva compound compound that functions as dual or plural expression

echo word formation word formation process in which some or all of the sounds

of the base word are repeated

elative case case that expresses ‘outward movement’

ellipsis omission of a retrievable part of a linguistic expression

elliptical construction construction in which one or more elements are not

expressed overtly

enclisis the attachment of a clitic to a preceding host word

endocentric compound compound of which one of the constituents functions as

its head

equative form of an adjective that expresses the meaning ‘to the same degree’

ergative case case marking of the Subject argument for verbs with more than one

argument in an absolutive-ergative system

established word word that belongs to the lexical convention of a language

evaluative morphology morphological expression of evaluative meanings such as

endearment or aVection

312 glossary

evidential morphological expression of the source or the reliability of the infor-

mation expressed in the relevant sentence

exclusive plural person marking for speaker and non-addressee

exemplar-based model model of linguistic knowledge in which individual com-

plex words and tokens thereof are represented as exemplars in lexical memory

exocentric compound compound without a constituent that functions as its head

explicit transposition the change of word class of a set of base words marked

through the addition of overt morphological material to these words

exponence the expression of grammatical categories through morphological

markers

exponent the morphological element that expresses a grammatical category

extended exponence the marking of a grammatical category by more than one

morphological element

external change change caused by external factors such as language contact

family size the number of derivationally related words of a base word

Wnite verbal form that bears morphological markers for categories such as

person and number, and can function as the only verbal form in a regular

sentence

formally complex word word that behaves as a complex word although there is

no corresponding semantic complexity

free morpheme morpheme that can function as a word of its own

frequency eVect the eVect of frequency of use of a linguistic item on the speed

with which that item is processed

full reduplication reduplication whereby the base word is copied completely

fusion the simultaneous expression of more than one grammatical category

through one morphological element

fusion rule rule that fuses grammatical categories into one node in a morpho-

syntactic representation

fusional language language with a tendency for fusion in its inXectional system

gapping the omission of a linguistic element under identity with another

occurrence of that linguistic element in the linguistic context

gender division of nouns into sets with diVerent inXectional and/or syntactic

behaviour

genitive case case that marks the dependency of a noun or noun phrase on

another noun

genitivus objectivus genitive case that expresses that the noun marked as such

functions as the object argument of another noun

glossary 313

genitivus subjectivus genitive case that expresses that the noun marked as such

functions as the subject argument of another noun

gerund verbal form with a nominal role

gerundive nominal verbal form with the (partial) distribution of a noun

Gesamtbedeuting the common denominator of the meaning of a polysemous

linguistic element

Goal semantic role of the entity towards which the action expressed by the verb is

directed

government an asymmetric structural relation between two linguistic elements

where one element determines properties of the other one

grammatical aspect expression of aspect by means of grammatical categories

grammatical function the function of an argument such as subject or direct object

grammatical word 1. a word that is one of the forms of a lexeme in an inXectional

paradigm; 2. a linguistic unit whose parts cannot be manipulated by syntactic

rules and thus exhibits lexical integrity; 3. a function word

grammaticalization the historical process in which lexical items become gram-

matical words

hapax (also called hapax legomenon) a word that occurs only once in a corpus

head the element in a construction that determines the properties of that con-

struction

head marking the marking of the relationship between a head noun and a

dependent noun on the head noun

head operation morphological operation in which a grammatical property of a

construction is marked morphologically on the head of that construction

honoriWc morphological marker that encodes the social relationship between

speaker and addressee

hypocoristic a pet-name, often used as a proper name to express aVection

iconicity the phenomenon that the meaning of a linguistic expression is reXected

by its formal structure

illative case case that expresses the meaning ‘into’

imperative verbal form that expresses a command

imperfect learning incomplete acquisition of a language that may lead to

language change

imperfective aspect aspect that expresses the situation as ongoing

implicational universal universal in the form of an implicational relation between

the presence of two diVerent properties

implicit transposition the change of word class of a set of words without overt

morphological marking

314 glossary

inclusive plural person marking for speaker and addressee

indicative mood of verb for expressing statements

inessive case case that expresses the notion ‘inside of ’

inWnite synonym of ‘non-Wnite’; verbal form that does not express information

for tense, number, and person

inWnitive non-Wnite form of a verb with nominal properties

inWx aYx that is inserted within a stem

inXection the morphological expression of morpho-syntactic categories on words

inXectional class a set of words with the same inXectional morphology

inXectional ending morphological marking of inXectional properties at the end of

a word

inXectional homonymy synonym of ‘syncretism’

inXectional paradigm the structured set of inXectional forms of a word

inherent case a synonym of ‘semantic case’

inherent inXection inXection of a word that is not required by its syntactic context

inheritance the transfer of properties of a word to the derived word of which it

forms the base

inheritance tree representation in the lexicon of the shared properties of sets of

words

input constraint requirement that a morphological process imposes on the kind

of input words it can take

interWx synonym of ‘linking element’

interlinear morphemic translation morpheme-by-morpheme translation of a lin-

guistic expression

internal change change caused by the dynamics of language use

internal modiWcation pattern of alternation of sounds in the roots of inXectionally

or derivationally related word(form)s

interpretation the assignment of meaning to a linguistic construction

invisible hand theory theory of language change modelled after Adam Smith’s

theory of the economic market: individuals do not behave in a goal-oriented

manner in changing languages but yet language change is the end result of the

behaviour of individuals

irrealis form of verb that indicates that something has not happened or is

unlikely to happen

isolating language language that does not make use of morphology

Item-and-Arrangement morphology model of morphological structure in which a

complex word consists of a sequence of concatenated morphemes

Item-and-Process morphology model of morphological structure in which each

complex word is the output of one or more morphological processes such as

aYxation and internal modiWcation

glossary 315

labelling function the function of words to label entities

language contact contact between speakers of two diVerent languages with eVects

on the lexicon and/or the grammar of one or both languages

levelling diachronic process in which a morphological process becomes more

regular

lexeme the abstract unit that stands for the common properties of all the forms

of a word

lexeme-based morphology morphology that takes lexemes as its inputs

lexeme formation the formation of new lexemes

lexical aspect inherent aspectual property of a lexical item

Lexical Conceptual Structure the representation of the meaning of a lexeme in

terms of conceptual structure

lexical convention the set of existing words of a language that functions as a norm

for language use

lexical decision task in a psycholinguistic experiment in which subjects are asked

to decide on a property of words presented in the experiment

lexical innovation the creation of new words to express new concepts

Lexical Integrity the principle that the constituents of a complex word cannot be

operated upon by rules of syntax

lexical morpheme synonym of ‘free morpheme’

lexical norm the set of existing words of a language that functions as a norm for

language use

Lexical Phonology theory of the interaction between morphology and

phonology that states that morphological and phonological rules apply in

tandem

lexicalization the phenomenon that lexical items acquire unpredictable

properties

linking element meaningless element between two constituents of a complex

word

linking rule rule that predicts the predicate-argument structure of a verb from its

lexical-conceptual structure

listeme synonym of ‘lexical unit stored in the lexicon’

local case case that expresses a spatial notion

local markedness the reversal of what the marked property is in a speciWc local

context

markedness 1. the phenomenon that a particular property of a lexical item is

expressed (marked) morphologically; 2. the phenomenon that a property of a

lexical item is not expected by some rule or principle

meaning that what is expressed by a linguistic unit or construction

316 glossary

mental lexicon the knowledge of the lexicon of a language as represented in the

individual minds of speakers

metaphony synonym of ‘Umlaut’

metonymy use of a word to express a concept that is associated with that word in

its literal sense

middle verb verb with a reXexive or passive meaning

middle voice voice in which the participant denoted by the subject argument is

also the beneWciary of the action

mixed category category of words that exhibit distributional behaviour of more

than one major lexical category

mixed inXection simpliWed form of strong inXection

mood the degree of actuality of an event as expressed by inXectional forms

morph the phonetic form of a morpheme

morpheme the minimal meaning-bearing unit of a language

morpheme-based morphology morphology that takes morphemes as its basic

building-blocks

morphololexical rule phonological rule whose application is governed by

morphological or lexical information

morphome a set of a morphemes in complementary distribution that functions as

stems in an inXectional or derivational paradigm

morphophonology 1. the use of phonology for morphological purposes; 2. phon-

ology as determined in its application by morphological information

morpho-syntactic category category of words that share a property with mor-

phological and/or syntactic eVects

multifunctionality the phenomenon that words can be used for diVerent syntactic

categories without overt diVerential marking

neo-classical compound compound that contains one or more Latin or

Greek roots

neo-classical word formation word formation that makes use of Latin or Greek

roots

neologism new word that does not yet belong to the lexical norm of a language

neuter gender gender of words that are neither feminine nor masculine

nominalization morphological process that turns words into nouns

nominative case case that marks the subject in a Nominative-Accusative system

Nominative-Accusative system system of case marking in which the subject

of intransitive verbs and transitive verbs receive the same nominative case

marking; the object of the transitive verb has an accusative case marking

non-cohering suYx suYx that forms a prosodic word of its own

glossary 317

non-concatenative morphology morphology that makes use of other processes

than aYxation or compounding to create new words or word forms

non-Wnite verbal form that does not express information for tense, number, and

person

No-Phrase Constraint constraint that forbids phrases forming parts of complex

words

noun incorporation operation in which nouns are made part of complex words

nucleus non-optional part of the syllable where vowels and syllabic consonants

appear

oblique case non-nominative structural case

obstruent consonant articulated with high degree of obstruction (stop

or fricative)

onset the consonantal beginning of a syllable before its nucleus

output constraint constraint on the output form of a linguistic expression

overcharacterization redundant morphological expression of a semantic property

of a word

overgeneralization incorrect generalization of a rule to more cases than the rule

should apply to

pan-European lexicon the lexical stock that is shared by most European

languages

Panini’s Principle principle that states that when two rules are both applicable to

a linguistic unit, the one with the most speciWc requirements takes priority, and

blocks the application of the more general rule

paradigm 1. structured set of inXectional forms of a word; 2. set of words that

contain the same word as a building block

paradigmatic levelling linguistic change in which words belonging to the same

paradigm become more similar as to their phonetic form

paradigmatic word formation word formation through extension of a

pattern based on paradigmatic relationships between established words

paradigmatically governed allomorphy choice of allomorph determined by a

paradigmatically related form

parasynthetic word formation word formation by simultaneous attachment of

both a preWx and a suYx

partial reduplication reduplication whereby only part of the base word is copied

participle non-Wnite form of a verb with both verbal and adjectival properties

particle verb phrasal combination of a particle and a verb that forms a lexical

unit and functions as a complex verb

partitive case case that expresses the notion ‘part of ’

318 glossary

passive voice voice in which the agent role is not linked to the grammatical

function of subject, and the grammatical function of the subject may then be

linked to the Patient role

passivization the operation of turning a verb into its passive counterpart

Patient semantic role of the participant of an action that undergoes that action

paucal number morphological form that expresses that a few entities are involved

percolation the transfer of features from a node n to a node that dominates n

perfect form of tense that indicates the continuing relevance of an event in the

past

perfective aspect aspect that presents an action as completed

periphrasis the use of more than one word to express a meaning that, in another

part of the morphology of a language, or in other languages, is expressed by

one complex word

person morpho-syntactic category that expresses speaker, addressee, or another

entity

phonological word a prosodic constituent consisting of minimally one syllable; a

grammatical word that is not a function word consists of one or more prosodic

words

phrasal aYx aYx that attaches to a phrase

phrasal verb verbal complex with phrasal status and the function of a complex verb

pidgin language simpliWed form of a language used in contact between speakers

who do not command a common language

pluralis majestatis plural form used for singular persons in order to express high

social status or respect

polymorphemic consisting of more than one morpheme

polysemy the phenomenon that a word has more than one meaning and these

meanings are related

polysynthetic language language that makes extensive use of word formation

processes for the construction of linguistic expressions

portmanteau morph morph that is the expression of more than one

morpho-syntactic category

position class morphology type of morphology in which there is a speciWc slot in

the complex morphological structure of words for the expression of each

morpho-syntactic category

positive degree the neutral degree form of adjectives

postlexical aYx aYx that is attached to a word postlexically

praesens historicum present form of verb used in story-telling in order to enhance

the liveliness of the performance

Predicate Argument Structure the speciWcation of the meaning of a predicate and

of the participants in the event denoted by the predicate that receives obliga-

tory syntactic expression

glossary 319

predicational aspect aspect expressed by the predicate of a clause

preWx aYx that is attached at the left edge of a stem

preverb element that appears in front of a verb, and forms a linguistic unit with

that verb

priming the presentation of information to a subject in a psycholinguistic

experiment that inXuences the performance on the task to be performed

proclisis the attachment of a clitic at the left edge of the host word

Pro-Drop parameter the parameter of variation between languages with respect

to the overt expression of the subject of a sentence

productivity property of a morphological process: a process is productive if it can

be applied to form new (forms of ) words

prosodic morphology the set of morphological processes that are deWned in terms

of prosodic categories

prosodic structure the structure of a linguistic expression deWned in prosodic

categories such as the syllables, the foot, and the prosodic word

prosodic word a synonym of ‘phonological word’

prosody the organization of phonological properties of linguistic expressions in

suprasegmental constituents such as syllable, tones, and intonation patterns

progressive aspect aspect that expresses the ongoing nature of an event

purist a language user who wants to avoid the borrowing of linguistic expressions

in his native language from other languages

qualifying adjective adjective that denotes a quality of the noun that it modiWes

radial category a semantic category with one or more core meanings from which

other meanings derive

readjustment rule in the framework of Distributed Morphology a rule that

performs a phonological operation on a restricted set of lexical items that are

diacritically marked for that rule

realis form of verb that indicates that something has happened, or happens

realization rule rule that spells out one or more morpho-syntactic properties of a

lexeme

realizational morphology morphological theory that conceives of morphological

processes as the spell-out rules of morpho-syntactic properties of lexemes

reanalysis the assignment of another structure than the original one to a

linguistic expression

recategorization the function of morphology to assign another syntactic category

to a linguistic expression

Recipient the semantic role for the participant in an event that receives

something

320 glossary

redundancy rule rule that states regularities in the properties of listed lexical units

reduplicant the part of a reduplicated word that is not the base

reduplication the morphological operation in which a new word (form) is created

by copying a word or a part thereof, and aYxing that copy to the base

relational adjective adjective that denotes a relation between the head noun that

it modiWes and some other entity evoked by the adjective

relative frequency the frequency of a complex word compared to that of its base

response latency the time interval between the appearance of information that is

to be processed and the reaction of the subject

result noun noun that expresses the result of an action

rhyme the part of the syllable after the consonantal onset and consisting of

nucleus and coda

Right-hand Head Rule rule that states that the right constituent of a complex

word functions as its head, and thus determines the properties of the complex

word

root the stem of a simplex word

root-and-pattern morphology form of non-concatenative morphology in which

the consonants of the root are intertwined with abstract patterns of consonants

and vowels with a grammatical value

rule of referral rule that expresses a generalization concerning the forms in cells

of a paradigm by referring to other cells

rule-governed creativity the ability of language users to coin new linguistic

expressions based on their knowledge of the rules of the language

Saussurean paradox the paradox that a language is a system located in

individuals and yet a system that must be shared by its users and hence belongs

to a community of speakers

schema pattern that expresses the common properties of a set of linguistic units

such as words

scope the domain of a semantic operator

semantic case case used for the expression of a semantic notion

separable complex verb phrasal lexical unit consisting of a verb and a non-verbal

element that functions like a complex verb

serial verb coordination of two or more verbs that function as one complex verb

in a sentence

shifter linguistic unit such as a pronoun whose reference depends on the situation

in which and the speaker by whom it is uttered

simplex word a word without internal morphological structure

single system model of linguistic knowledge or processing that does not make a

distinction between rules and units of representation

glossary 321

singulative morphological marker for the property ‘singular’

sound symbolism the use of sounds (combinations) to express meaning in a

partially systematic way

spell out rule rule that speciWes the phonological realization of (combinations of )

morpho-syntactic properties of words

split ergativity the phenomenon that in one language both a nominative-

accusative and an absolute-ergative system are used for case marking

split morphology the hypothesis that inXection and derivation are to be

accounted for in separate modules of the grammar of natural languages

stem the form of a word that is obtained by removing its inXectional endings and

that functions as the base form for morphological processes

stem allomorphy variation in the phonological shape of a stem that is condi-

tioned by non-phonological factors

stratal constraint constraint on the possible bases of a morphological process

that refers to a particular stratum of the lexicon

stratum subset of words in the lexicon with speciWc properties that often reXect

the historical origin of that set of words

strong inXection inXection with maximal distinctness between the cells of

a paradigm

strong lexicalism the claim that inXection and derivation should be accounted for

in the same, pre-syntactic, module of the grammar

structural case case required by a syntactic context

subcategorization the division of the members of a word class in subsets with

speciWc properties

subjunctive form of the verb that expresses subordination, unreality, or desire

suYx aYx that is attached at the right edge of a stem

superlative degree form of an adjective that expresses that a property is present to

the highest degree possible

suppletion the phenomenon that forms of diVerent roots Wll the cells of a

paradigm

supraWx aYx of a suprasegmental nature such as tone

suprasegmental a phonological property of more than one sound segment

switch reference the phenomenon that a referential diVerence between subjects of

consecutive clauses is marked morphologically

syllable the prosodic organization of sounds into a higher-order unit, with a

vowel or syllabic consonant as its nucleus

syllable structure the internal structure of a syllable: onset and rhyme, with the

rhyme consisting of nucleus and coda

syncretism the phenomenon that diVerent cells of a paradigm are Wlled by the

same phonological word form

322 glossary

syntactic valency the number of arguments that a word can license

syntagmatic relationship relationship between linguistic units that are combined

in a linguistic expression; relationship on the horizontal axis of language

structure

synthesis the phenomenon that complex information is expressed by means of

complex words instead of by syntactic constructions

synthetic compounding word formation in which compounding and derivation

are combined

synthetic form the opposite of ‘analytic form’: situation in which complex

information is expressed by one, complex, word

systematization form of morphological change by which the situation of

‘one form—one meaning’ is obtained for the morphological encoding of a

particular semantic domain

telic aspect aspect that expresses that the endpoint of an action is implied

template a structural scheme for a set of linguistic units

tense the grammatical expression of time notions

thematic role semantic role of a participant in an event

thematic vowel vowel that is used to form a stem from a root

TMA system system for the expression of tense, mood, and aspectual categories

token frequency the number of tokens of a linguistic unit or a class of linguistic

units in a corpus

tonal morpheme tone pattern with the function of a morpheme, that is, expressing

a sound–meaning correspondence

transcategorial construction morphological construction with properties of more

than one word class, such as the inWnitive and the participle

transWx an abstract aYx expressed by the pattern of vowels of a root

transparency the presence of a systematic form–meaning correspondence in a

morphologically complex form

transposition the change of word class of a set of base words

trial morphological form for the number ‘three’

truncation the removal of phonological material as part of a word formation

process

type coercion the semantic interpretation imposed on a base word by a

morphological process

type frequency the number of types of a class of linguistic units in a corpus

umlaut a synonym of ‘vowel mutation’ and ‘metaphony’: pattern of vowel

alternation in the roots of inXectionally or derivationally related word(form)s

which is historically the eVect of assimilation

glossary 323

underlying form the phonological representation of a word on the basis of which

its diVerent phonetic forms can be computed by means of phonological rules

unidirectionality the hypothesis that grammaticalization processes are

unidirectional and cannot be reversed, and always lead from less grammatical

to more grammatical forms

univerbation linguistic change in which (parts of) phrases become one word

universal generalization concerning properties of natural languages

vagueness the phenomenon that the meaning of a linguistic expression is highly

abstract or underspeciWed

verbalization the formation of verbs through a morphological process

voice pattern of linking between semantic roles and grammatical functions such

as active and passive

vowel gradation synonym of ‘Ablaut’ and ‘apophony’: pattern of vowel

alternation in the roots of inXectionally or derivationally related word(form)s

vowel harmony the phenomenon that all vowels of a (prosodic) word have to

share a certain property, for instance frontness or backness

vowel mutation a synonym of ‘Umlaut’ and ‘metaphony’: pattern of vowel alter-

nation in the roots of inXectionally or derivationally related word(form)s

which is historically the eVect of assimilation

weak inXection inXection with reduced formal distinctness of the diVerent cells of

a paradigm

Word-and-Paradigm morphology theoretical model of inXection that takes the

lexeme and its paradigm of cells as the starting point for the analysis of

inXectional systems

word creation the creation of new words by other means than regular morpho-

logical processes

word family set of words that share the same root

word form one of the inXectional forms of a lexeme

word manufacturing a synonym of ‘word creation’

word recognition the matching of linguistic information presented auditively or

visually with a word in lexical memory

world knowledge the non-linguistic knowledge that plays a role in interpreting

words

zero-ending an abstract suYx that represents a set of inXectional properties but

has no overt phonological form

zero-morpheme an abstract morpheme that represents morpho-syntactic or

semantic information but has no overt phonological form

324 glossary

References

BLM Booij, G., Lehmann, Ch., and Mugdan, J. (eds.), Morphology. An Inter-

national Handbook of Inflection andWord Formation (Berlin: De Gruyter),

i (2000), ii (2004).

JL Journal of Linguistics.

Lg Language. Journal of the Linguistic Society of America.

NLLT Natural Language and Linguistic Theory.

YoM Yearbook of Morphology, edited by Geert Booij and Jaap van Marle.

Dordrecht: Foris (1988–90), Dordrecht: Kluwer Academic Publishers

(1991– ).

Ackerman, F., and Goldberg, A. (1996). ‘Constraints on Adjectival Past Parti-

ciples’, in A. E. Goldberg (ed.), Conceptual Structure, Discourse and Language,

17–30. Stanford: CSLI.

Aitchison, J. (1987). Words in the Mind: An Introduction to the Mental Lexicon.

Oxford: Blackwell [2002, 3rd revised edition].

—— (2003). ‘Psycholinguistic Perspectives on Language Change’, in Joseph and

Janda 2003: 736–43.

—— (2004). ‘Speech Perception and Production’, BLM ii, art. 163.

Anderson, S. R. (1985). ‘Inflectional Morphology’, in Shopen 1985: 150–201.

—— (1992). A-Morphous Morphology. Cambridge: Cambridge University Press.

—— (2005). Aspects of the Theory of Clitics. Oxford: Oxford University Press.

Andersson, E. (1994). ‘Swedish’, in Konig and van der Auwera 1994: 271–312.

Aronoff, M. (1976). Word Formation in Generative Grammar. Cambridge, Mass.:

MIT Press.

—— (1980). ‘Contextuals’, Lg 56: 744–58.

—— (1994). Morphology by Itself: Stems and Inflectional Classes. Cambridge

Mass.: MIT Press.

—— and Fuhrhop, N. (2002). ‘Restricting Suffix Combinations in German and

English: Closing Suffixes and the Monosuffix Constraint’, NLLT 20: 451–90.

Austin, P. (1981). A Grammar of Diyari, South Australia. Cambridge: Cambridge

University Press.

Ayres, G. (1983). ‘The Antipassive ‘‘Voice’’ in Ixil’, International Journal of Ameri-

can Linguistics 49: 20–45.

Baayen, R. H. (1992). ‘Quantitative Aspects of Morphological Productivity’, YoM

1991, 109–50.

Baayen, R. H. (1993). ‘On Frequency, Transparency, and Productivity’, YoM

1992, 181–208.

—— (2003). ‘Probabilistic Approaches to Morphology’, in R. Bod, J. Hay, and

S. Jannedy (eds.), Probabilistic Linguistics, 229–87. Cambridge, Mass., and

London: MIT Press.

—— Piepenbrock, R., and Gulikers, L. (1995). The CELEX Lexical Database.

Phildelphia: Linguistic Data Consortium, University of Philadelphia (CD-rom).

—— Burani, C., and Schreuder, R. (1997). ‘Effects of Semantic Markedness in

the Processing of Regular Nominal Singulars and Plurals in Italian’, YoM 1996,

13–34.

—— Dijkstra, T., and Schreuder, R. (1997). ‘Singulars and Plurals in Dutch:

Evidence for a Parallel Dual Route Model’, Journal of Memory and Language 36:

94–117.

—— Schreuder, R., de Jong, N. and Krott, A. (2002). ‘Dutch Inflection: The

Rules that Prove the Exception’, in Nooteboom et al. 2002: 61–92.

Baker, M. (1988). Incorporation: A Theory of Grammatical Function Changing.

Chicago and London: University of Chicago Press.

—— (1996). The Polysynthesis Parameter. New York and Oxford: Oxford Univer-

sity Press.

—— (2001). The Atoms of Language. New York: Basic Books.

Bakker, P. (2003). ‘Pidgin Inflectional Morphology and its Implications for Creole

Morphology’, YoM 2003, 3–35.

Barlow, M., and Ferguson, C. (eds.) (1988). Agreement in Natural Language.

Approaches, Theories, Descriptions. Stanford, Calif.: CSLI.

Bauer, L. (2000). ‘Word’. BLM vol 1: 247–57.

—— (2001). Morphological Productivity. Cambridge: Cambridge University Press.

Bauer, W. (1993). Maori. London and New York: Routledge.

Beard, R. (1991). ‘Decompositional Composition: The Semantics of Scope Ambi-

guities and ‘‘Bracketing Paradoxes’’ ’, NLLT 9: 195–229.

Becker, T. (1990a). Analogie und morphologische Theorie. Munich: Wilhelm Fink

Verlag.

—— (1990b). ‘Do Words Have Heads?’, Acta Linguistica Hungarica 40: 5–17.

Benveniste, E. (1967). ‘Fondements syntaxiques de la composition nominale’,

published as Chapter 11 of Problemes de linguistique generale. Paris: Gallimard,

1974.

Berent, I., Marcus, G. F., Shimron, J., and Gafos, A. I. (2002). ‘The Scope of

Linguistic Generalizations: Evidence from Hebrew Word Formation’, Cognition

83: 113–39.

Berko, J. (1958). ‘The Child’s Learning of English Morphology’,Word 14: 150–77.

326 references

Bhatia, T. K. (1993). Punjabi: A Cognitive-Descriptive Grammar. London and New

York: Routledge.

Blake, B. (1994). Case. Cambridge: Cambridge University Press.

Bleser, R. de, and Bayer, J. (1988). ‘On the Role of Inflectional Morphology in

Agrammatism’, in Hammond and Noonan 1988: 45–69.

Blom, C., and Booij, G. E. (2003). ‘The Diachrony of Complex Predicates in

Dutch: A Case Study in Grammaticalization’, Acta Linguistica Hungarica 50:

61–91.

Bloomfield, L. (1935), Language. London: George Allen & Unwin Ltd.

Booij, G. E. (1977). Dutch Morphology: A Study of Word Formation in Generative

Grammar. Dordrecht: Foris.

—— (1985). ‘Coordination Reduction in Complex Words: A Case for Prosodic

Phonology’, in H. van der Hulst and N. Smith (eds.) Advances in Non-Linear

Phonology, 143–60. Dordrecht: Foris.

—— (1986). ‘Form and Meaning in Morphology: The Case of Dutch ‘‘Agent’’

Nouns’, Linguistics 24: 503–17.

—— (1988). ‘The Relation between Inheritance and Argument Linking: Deverbal

Nouns in Dutch’, in Everaert et al. 1988: 57–74.

—— (1992). ‘Morphology, Semantics, and Argument Structure’, in I. Roca (ed.),

Thematic Structure: Its Role in Grammar, 47–63. Berlin and New York: Foris.

—— (1994). ‘Against Split Morphology’, YoM 1993, 27–50.

—— (1996a). ‘Inherent versus Contextual Inflection and the Split Morphology

Hypothesis’, YoM 1995, 1–16.

—— (1996b). ‘Cliticization as Prosodic Integration: The Case of Dutch’, The

Linguistic Review 13: 219–42.

—— (1997a). ‘Allomorphy and the Autonomy of Morphology’, Folia Linguistica

31: 25–56.

—— (1997b). ‘Autonomous Morphology and Paradigmatic Relations’, YoM 1997,

35–54.

—— (1997c). ‘Non-Derivational Phonology Meets Lexical Phonology’, in I. Roca

(ed.), Constraints and Derivations in Phonology, 261–88. Oxford: Clarendon

Press.

—— (1998). ‘Phonological Output Constraints inMorphology’, inW. Kehrein and

R. Wiese (eds.), Phonology and Morphology of the Germanic Languages, 143–63.

Tubingen: Niemeyer.

—— (2000). ‘The Phonology–Morphology Interface’, in L. Cheng and R. Sybesma

(eds.), The First Glot International State-of-the-Article Book, 287–306. Berlin:

Mouton de Gruyter.

—— (2002a). The Morphology of Dutch. Oxford: Oxford University Press.

—— (2002b). ‘Prosodic Restrictions on Stacking up Affixes’, YoM 2001, 183–202.

references 327

Booij, G. E. (2002c). ‘Constructional Idioms, Morphology, and the Dutch Lexi-

con’, Journal of Germanic Linguistics 14: 301–27.

—— (2005). ‘Context-dependent Morphology’, Lingue e Linguaggio 2: 163–78.

—— and Lieber, R. (2004). ‘On the Paradigmatic Nature of Affixal Semantics in

English and Dutch’, Linguistics 42: 327–57.

Borer, H. (1988). ‘On the Morphological Parallelism between Compounds and

Constructs’, YoM 1988, 45–66.

B�rjars, K., Vincent, N., and Chapman, C. (1997). ‘Paradigms, Periphrases, and

Pronominal Inflection’, YoM 1996, 155–80.

Bosch, P. (1983). Agreement and Anaphora: A Study of the Role of Pronouns in

Syntax and Discourse. New York: Academic Press.

Botha, R. P. (1984). Morphological Mechanisms: Lexicalist Analyses of Synthetic

Compounding. Oxford: Pergamon Press.

—— (1988). Form and Meaning in Word Formation: A Case Study of Afrikaans

Reduplication. Cambridge: Cambridge University Press.

Bresnan, J., and Mchombo, S. (1995). ‘The Lexical Integrity Principle: Evidence

from Bantu’, NLLT 13: 181–254.

Broselow, E. (2000). ‘Transfixation’, BLM i. 552–7.

Bybee, J. L. (1985). Morphology: A Study of the Relation between Meaning and

Form. Amsterdam: Benjamins.

—— (1988). ‘Morphology as Lexical Organization’, in Hammond and Noonan

1988: 119–42.

—— (1995). ‘Regular Morphology and the Lexicon’, Language and Cognitive

Processes 10: 425–55.

—— (2001). Phonology and Language Use. Cambridge: Cambridge University Press.

—— (2003). ‘Mechanisms of Change in Grammaticization: The Role of Frequency’,

in Joseph and Janda 2003: 602–23.

Cameron-Faulkner, T., and Carstairs-McCarthy, A. (2000). ‘Stem Alternants

as Morphological Signata: Evidence from Blur Avoidance in Polish Nouns’,

NLLT 18: 813–35.

Campbell, L. (2001). ‘What’s Wrong with Grammaticalization?’, Language

Sciences 23: 113–61.

Carstairs, A. (1987). Allomorphy in Inflexion. London: Croom Helm.

—— (1988). ‘Some Implications of Phonologically Conditioned Suppletion’, YoM

1988, 67–94.

Carstairs-McCarthy, A. (1992). Current Morphology. London and New York:

Routledge.

—— (1993). ‘Morphology without Word-Internal Constituents: A Review of

Anderson (1992)’, YoM 1992, 209–34.

328 references

—— (1994). ‘Inflection Classes, Gender, and the Principle of Contrast’, Lg 70:

737–88.

Chomsky, N., and Halle, M. (1968). The Sound Pattern of English. New York:

Harper & Row.

Chung, S., and Timberlake, A. (1985). ‘Tense, Aspect, and Mood’, in Shopen

1985: 202–58.

Clahsen, H. (1999). ‘Lexical Entries and Rules of Language’, Behavorial and Brain

Sciences 22: 991–1060.

Clark, E. V. (1993). The Lexicon in Acquisition. Cambridge: Cambridge University

Press.

—— and Berman, R. A. (1984). ‘Structure and Use in the Acquisition of Word

Formation’, Lg 60: 542–90.

Clark and Berman (2004). ‘Morphology in First Language Acquisition’, BLM ii,

art. 165.

—— and Clark, H. H. (1979). ‘When Nouns Surface as Verbs’, Lg 55: 767–811.

Closs Traugott, E. (2003). ‘Constructions in Grammaticalization’, in Joseph and

Janda 2003: 624–47.

Coates, R. (2000). ‘Exponence’, BLM i. 616–30.

Comrie, B. (1976). Aspect. Cambridge: Cambridge University Press.

—— (1981). Language Universals and Linguistic Typology. Oxford: Blackwell.

—— (1984). Tense. Cambridge: Cambridge University Press.

—— (1985). ‘Causative Verb Formation and Other Verb-Deriving Morphology’,

in Shopen 1985: 309–48.

—— (2001). ‘Recipient Person Suppletion in the Verb ‘‘Give’’ ’ (lecture handout,

Berlin).

—— and Thompson, S. (1985). ‘Nominalization’, in Shopen 1985: 349–98.

Corbett, G. (1991). Gender. Cambridge: Cambridge University Press.

—— (2000). Number. Cambridge: Cambridge University Press.

—— and Fraser, N. M. (1993). ‘Network Morphology: a DATR Account of

Russian Nominal Inflection’, JL 29: 113–42.

Cowan, W., and Rakuƒan, J. (1985). Source Book for Linguistics. Amsterdam and

Philadelphia: Benjamins.

Croft, W. (1990). Typology and Universals. Cambridge: Cambridge University

Press.

Cruse, A. (2000). Meaning in Language: An Introduction to Semantics and Prag-

matics. Oxford: Oxford University Press.

Csat�, E. A., and Johanson, L. (1998). ‘Turkish’, in L. Johanson and E. A. Csato

(eds.), The Turkic Languages, 203–35. London and New York: Routledge.

Cutler, A. (ed.) (1982). Slips of the Tongue and Language Production, Linguistics

19, issue 7–8.

references 329

Cutler, A. Hawkins, J. A., and Gillegan, G. (1985). ‘The Suffixing Preference:

A Processing Explanation’, Linguistics 23: 723–59.

Cysouw, M. (2001). The Paradigmatic Structure of Person Marking. Ph.D. diss.

University of Nijmegen [revised edn. Oxford: Oxford University Press, 2003].

Dahl, O. (2004). The Growth and Maintenance of Linguistic Complexity. Amster-

dam/Philadelphia: John Benjamins.

Dell, F. and Selkirk, E. O. (1978). ‘On a Morphologically Governed Vowel

Alternation in French’, in S. J. Keyser (ed.), Recent Transformational Studies in

European Languages, 1–51. Cambridge, Mass.: MIT Press.

Dimmendaal, G. (1983). The Turkana Language. Dordrecht: Foris.

—— (2002). ‘Morphology’, in B. Heine and D. Nurse (eds.),African Languages: An

Introduction, 161–93. Cambridge: Cambridge University Press.

DiSciullo, A. M., andWilliams, E. (1987).On the Definition ofWord. Cambridge,

Mass.: MIT Press.

Dixon, R. M. W. (1977). A Grammar of YidiÆ. Cambridge: Cambridge University

Press.

—— (1994). Ergativity. Cambridge: Cambridge University Press.

—— and Aikhenvald, A. Y. (eds.) (2000). Changing Valency: Case Studies in

Transitivity. Cambridge: Cambridge University Press.

—— (eds.) (2002). Word. A Cross-Linguistic Typology. Cambridge: Cambridge

University Press.

Downing, P. (1977). ‘On the Creation and Use of English Compound Nouns’, Lg

53: 810–42.

Dowty, D. (1991). ‘Thematic Proto-Roles and Argument Selection’, Lg 67:

547–619.

Dressler, W. U., and Barbaresi, L. Merlini (1994). Morphopragmatics: Diminu-

tives and Intensifiers in Italian, German, and Other Languages. Berlin and New

York: Mouton de Gruyter.

Dyk, S. (1988). ‘Oer it foarheaksel witte- (withoe-) en syn syntaktysk komof ’, in

S. Dyk and G. de Haan (eds.), Wurdfoarried en Wurdgrammatika, 21–44.

Ljouwert: Fryske Akademy.

—— (1997). Noun Incorporation in Frisian. Ljouwert: Fryske Akademy.

Eisenberg, P. (1994). Grundriss der deutschen Grammatik. Stuttgart and Weimar:

Verlag J. B. Metzler.

Elbers, L. (1988). ‘New Names for Old Words: Related Aspects of Children’s

Metaphors and Word Compounds’, Journal of Child Language 5: 591–617.

Everaert, M., Evers, A., Huybregts, R., and Trommelen, M. (eds.) (1988).

Morphology and Modularity: In Honour of Henk Schultink. Dordrecht: Foris.

Fabb, N. (1988). ‘English Suffixation is Constrained Only by Selectional

Restrictions’, NLLT 6: 527–39.

330 references

Fleischmann, S. (1977). Cultural and Linguistic Factors in Word Formation: An

Integrated Approach to the Development of the Suffix-age. Berkeley, Calif.: Uni-

versity of California Press.

Fortescue, M. (1984). West-Greenlandic. Beckenham: Croom Helm.

Frauenfelder, U. H., and Schreuder, R. (1992). ‘Constraining Psycholinguistic

Models of Morphological Processing and Representation: The Role of

Productivity’, YoM 1991, 165–84.

Fromkin, V. (ed.) (1973). Speech Errors as Linguistic Evidence. The Hague and

Paris: Mouton.

Goldberg, A., and Ackerman, F. (2001). ‘The Pragmatics of Obligatory

Adjuncts’, Lg 77: 798–814.

GonzÆlez, P. (2003). Aspects on Aspect: Theory and Applications of Grammatical

Aspect in Spanish. Utrecht: LOT.

Greenberg J. (1963). ‘Some Universals of Grammar with Particular Reference to

the Order of Meaningful Elements’, in id. (ed.), Universals of Language, 73–113.

Cambridge Mass.: MIT Press. [19662].

Grimes, B. (ed.) (2003). Ethnologue: Languages of the World. Dallas, Tex.: SIL

[14th edn, also on the internet: www.ethnologue.com].

Grimshaw, J. (1990). Argument Structure. Cambridge, Mass.: MIT Press.

G�rel, A. (1999). ‘Decomposition: To What Extent? The Case of Turkish’, Brain

and Language 68: 218–24.

Hagemeijer, T. (2001). ‘Underspecification in Serial Verb Constructions’, in

N. Corver and H. van Riemsdijk (eds.), Semi-Lexical Categories, 415–51. Berlin

and New York: Mouton de Gruyter.

Hall, C. J. (1988). ‘Integrating Diachronic and Processing Principles in Explaining

the Suffix Preference’, in Hawkins 1988: 321–59.

—— (1991). Morphology and Mind: A Unified Approach to Explanation in

Linguistics. London and New York: Routledge.

Hall, T. A. and Kleinhenz, U. (eds.) (1999). Studies on the Phonological Word.

Amsterdam/Philadelphia: Benjamins.

Halle, M. (1992). ‘The Latvian Declension’, YoM 2001, 33–47.

—— and Marantz, A. (1993). ‘Distributed Morphology’, in K. Hale and

S. J. Keyser (eds.), The View from Building 20: Essays in Linguistics in Honor of

Sylvain Bromberger, 111–76. Cambridge Mass.: MIT Press.

Hammond, M., and Noonan, M. (eds.) (1988). Theoretical Morphology:

Approaches in Modern Linguistics. San Diego: Academic Press.

Harada, S. I. (1976). ‘Honorofics’, in M. Shibatani (ed.), Japanese Generative

Grammar, 499–561. New York: Academic Press.

Harley, H., and Noyer, R. (2003). ‘Distributed Morphology’, in L. Cheng and

R. Sybesma (eds.), The Second GLOT International State-of-the-Article Book,

463–96. Berlin: Mouton de Gruyter.

references 331

www.ethnologue.com

Harris, A. (2006). ‘In Other Words: External Modifiers in Georgian’, Morphology

16 (2).

Haspelmath, M. (1996). ‘Wordclass-Changing Inflection and Morphological

Theory’, YoM 1995, 43–66.

Hawkins, J. A. (ed.) (1988). Explaining Language Universals. Oxford: Blackwell.

—— and Cutler, A. (1988). ‘Psycholinguistic Factors in Morphological

Asymmetry’, in Hawkins 1988: 280–317.

Hay, J. (2001). ‘Lexical Frequency in Morphology. Is Everything Relative?’,

Linguistics 39: 1041–70.

—— (2002). ‘From Speech Perception to Morphology: Affix Ordering Revisited’,

Lg 78: 527–55.

—— and Baayen, R. H. (2002). ‘Parsing and Productivity’, YoM 2001, 203–36.

Heine, B. (2003). ‘Grammaticalization’, in Joseph and Janda 2003: 575–601.

—— and Kuteva, T. (2002). World Lexicon of Grammaticalization. Cambridge:

Cambridge University Press.

—— Claudi, U., and H�nnemeyer, F. (1991). Grammaticalization: A Conceptual

Framework. Chicago and London: University of Chicago Press.

Hoeksema, J. (1985). Categorial Morphology. New York: Garland Press.

Hoekstra, T. (1986). ‘Deverbalization and Inheritance’, Linguistics 24: 549–84.

—— and van der Putten, F. (1988). ‘Inheritance Phenomena’, in Everaert et al.

1988: 163–86.

Hopper, P. J., and Closs Traugott, E. (1993). Grammaticalization. Cambridge:


Hyman, L. (2003). ‘Suffix Ordering in Bantu: A Morphocentric Approach’, YoM

2003, 245–83.

Inkelas, S. (1993). ‘Nimboran Position Class Morphology’, NLLT 11: 559–624.

Jackendoff, R. S. (1975). ‘Semantic and Morphological Regularities in the

Lexicon’, Lg 51: 639–71.

—— (1983). Semantics and Cognition. Cambridge, Mass.: MIT Press.

—— (1997). The Architecture of the Language Faculty. Cambridge, Mass.: MIT

Press.

—— (2002). Foundations of Language. Oxford: Oxford University Press.

Jaeger, J. J., Lockwood, A. H., Kemmerer, D. L., Van Valin, R. D., Murphy,

B. W., and Khalak, H. G. (1996). ‘A Positron Emission Tomographic Study of

Regular and Irregular Verb Morphology in English’, Lg 72: 451–97.

Jakobson, R. (1936). ‘Beitrag zur allgemeinen Kasuslehre’, reprinted in Roman

Jakobson: Selected Writings, ii. Word and Language, 23–71. The Hague and

Paris: Mouton.

Janssen, D., Roelofs, A., and Levelt, W. J. M. (2002). ‘Inflectional Frames in

Language Production’, Language and Cognitive Processes 17: 209–36.

332 references

Jong, N. H. de, Schreuder, R., and Baayen, R. H. (2000). ‘The Morphological

Family Size Effect andMorphology’,Language andCognitive Processes 15: 329–65.

Joseph, B.D., and Janda, R.D. (eds.) (2003).TheHandbook ofHistorical Linguistics.

Oxford: Blackwell.

Jurafsky, D. (1996). ‘Universal Tendencies in the Semantics of the Diminutive’, Lg

72: 533–78.

Kager, R. (1999). Optimality Theory. Cambridge: Cambridge University Press.

Kastovsky, D. (1986). ‘The Problem of Productivity in Word Formation’,

Linguistics 24: 585–600.

Katamba, F. (ed.) (2004). Morphology. Critical Concepts in Linguistics, 6 vols.

London: Routledge.

Keenan, E. (1976). ‘Towards a Universal Definition of Subject’, in C. Li (ed.),

Subject and Topic, 303–34. New York: Academic Press.

—— and Polinsky, M. (1998). ‘Malagasy (Austronesian)’, in Spencer and Zwicky

1998: 563–623.

Keller, R. (1994).On Language Change: The Invisible Hand in Language. London:

Routledge.

Kemenade, A. van, and Los, B. (2003). ‘Particles and Prefixes in Dutch and

English’, YoM 2003, 79–118.

Kerke, S. C. van de (1996). ‘Affix Order and Interpretation in Bolivian Quechua’,

Ph.D. diss. Univ. of Amsterdam.

Kiefer, F. (1992). ‘Compounding in Hungarian’, Rivista di Linguistica 4: 61–78.

Kiparsky. P. (1994). ‘Allomorphy or Morphophonology?’, in R. Singh (ed.),

Trubetzkoy’s Orphan: Proceedings of the Montreal Round-Table ‘Morphonology:

Contemporary Responses’, 13–62. Amsterdam and Philadelphia: Benjamins.

Klaiman, M. H. (1991). Grammatical Voice. Cambridge: Cambridge University

Press.

Klamer, M. (1998). A Grammar of Kambera. Berlin: Mouton de Gruyter.

K�nig, E., and van der Auwera, J. (eds.) (1994). The Germanic Languages. New

York: Routledge.

Koptjevskaja-Tamm, M. (1993). Nominalizations. London and New York: Rou-

tledge.

Kornfilt, J. (1997). Turkish. London and New York: Routledge.

Krott, A. (2001). ‘Analogy in Morphology: The Selection of Linking Elements in

Dutch Compounds’, Ph. D. diss. University of Nijmegen.

—— Schreuder, R., and Baayen, R. H. (2001). ‘Analogy in Morphology:

Modeling the Choice of Linking Morphemes in Dutch’, Linguistics 39: 51–93.

Kula, C. N. (2002). The Phonology of Verbal Derivation in Bemba. Utrecht: LOT.

Kutsch Lojenga, C. (1994).Ngiti: A Central-Sudanic Language of Zaire. Cologne:

Rudiger Koppe Verlag.

references 333

Lappe, S. (2003). ‘Monosyllabicity in Prosodic Morphology: the Case of Truncated

Personal Names in English’, YoM 2002, 35–86.

Lass, R. (1990). ‘How toDo Things with Junk: Exaptation in Language Evolution’,

JL 26: 79–102.

Lefebvre, C. (2003). ‘The Emergence of Productive Morphology in Creole

Languages: The Case of Haitian Creole’, YoM 2002, 35–80.

Lehmann, C. (1982). ‘Directions for Interlinear Morphemic Translations’, Folia

Linguistica 16: 199–224.

—— (2004). ‘Interlinear Morphemic Glossing’, BLM ii, art. 169.

Levelt, W. J. M. (1989). Speaking: From Intention to Articulation. Cambridge,

Mass.: MIT Press.

Levi, J. N. (1978). The Syntax and Semantics of Complex Nominals. New York:

Academic Press.

Lewis, G. L. (1967). Turkish Grammar. Oxford: Oxford University Press.

Lieber, R. (1980). On the Organization of the Lexicon. Ph.D. diss. MIT, published

by Garland Press, New York, 1994.

—— (1987). An Integrated Theory of Autosegmental Processes. Albany, NY:

SUNY Press.

—— (1989). ‘On Percolation’, YoM 1989, 95–138.

—— (1992). Deconstructing Morphology. Chicago: University of Chicago Press.

—— (2000). ‘Substitution of Segments and Features’, BLM i. 567–76.

McCarthy, J. J. (1981). ‘A Prosodic Theory of Non-Concatenative Morphology’,

Linguistic Inquiry 12: 373–418.

—— (1998). ‘Prosodic Morphology’, in Spencer and Zwicky 1998: 283–305.

—— and Prince, A. (1994). ‘Generalized Alignment’, YoM 1993, 79–154.

McWhorter, J. (1998). ‘Identifying the Creole Prototype: Vindicating a

Typological Class’, Lg 74: 788–818.

Maiden, M. (1992). ‘Irregularity as a Determinant of Morphological Change’, JL

28: 285–312.

Malouf, R. P. (2000). Mixed Categories in the Hierarchical Lexicon. Stanford

Calif.: CSLI.

Marantz, A. (1982). ‘Re Reduplication’ Linguistic Inquiry 13: 83–545.

Marchand, H. (1969). The Categories and Types of Present-Day English Word

Formation, 2nd edn. Munich: Beck.

Marcus, G. (2001). The Algebraic Mind: Integrating Connectionism and Cognitive

Science. Cambridge, Mass.: MIT Press.

Marle, J. van (1978). ‘Veranderingen in woordstructuur’, in G. A. T. Koefoed

and J. van Marle (eds.), Aspecten van taalverandering, 127–76. Groningen:

Wolters-Noordhoff.

—— (1985). On the Paradigmatic Dimension of Morphological Creativity. Dor-

drecht: Foris.

334 references

Martin, W., and Tops, G. A. J. (1984). Van Dale Groot Woordenboek Engels–

Nederlands. Utrecht: Van Dale Lexicografie.

Matthews, P. H. (1972). Inflectional Morphology: A Theoretical Study Based on

Aspects of Latin Verb Conjugation. Cambridge: Cambridge University Press.

Melc“uk, I. (2000). ‘Morphological Processes’, BLM i. 523–34.

Mithun, M. (1984). ‘The Evolution of Noun Incorporation’, Lg 60: 847–94.

—— (1999). The Languages of Native North America. Cambridge: Cambridge

University Press.

—— (2000). ‘Incorporation’, BLM i. 916–28.

Muysken, P. C. (1988). ‘Affix Order and Interpretation: Quechua’, in Everaert et al.

1988: 259–79.

Nau N. (2001). ‘Introduction’, in id. (ed.), Typological Approaches to Latvian.

Berlin: Akademieverlag (Sprachtypologie und Universalienforschung 54/3).

Nevis, J. (2000). ‘Clitics’. BLM Vol 1: 388–401.

Newmeyer, F. J. (1998). Language Form and Language Function. Cambridge,

Mass.: MIT Press.

Nichols, J. (1986). ‘Head-Marking and Dependent-Marking Grammar’, Lg 62:

56–119.

Nooteboom, S. G., Weerman, F., and Wijnen, F. (eds.) (2002). Storage and

Computation in the Language Faculty. Dordrecht: Kluwer.

Norde, M. (1997). ‘The History of the Genitive in Swedish: A Case Study in

Degrammaticalization’, Ph.D. diss. University of Amsterdam.

—— (2001). ‘Deflexion as a Counterdirectional Factor in Grammatical Change’,

Language Sciences 23: 231–64.

Olsen, S. (2000). ‘Composition’, BLM i. 898–916.

—— (2001). ‘Copulative Compounds: A Closer Look at the Interface between

Syntax and Morphology’, YoM 2000, 279–320.

Oniga, R. (1992). ‘Compounding in Latin’, Rivista di Linguistica 4: 97–116.

Payne, T. E. (1997). Describing Morpho-Syntax: A Guide for Field Linguists.

Cambridge: Cambridge University Press.

Pinker, S. (1999). Words and Rules. New York: Basic Books.

Plag, I. (1996). ‘Selectional Restrictions in English Suffixation Revisited’, Linguistics

34: 769–98.

—— (1999). Morphological Productivity: Structural Constraints in English

Derivation. Berlin and New York: Mouton de Gruyter.

—— (ed.) (2003a). ‘The Morphology of Creole Languages’, YoM 2002, 1–134.

—— (2003b). Word-Formation in English. Cambridge: Cambridge University

Press.

Plank, F. (1981). Morphologische (Ir-) Regularitaten. Tubingen: Narr.

Rainer, F. (2000). ‘Produktivitatsbeschrankungen’, BLM i. 877–85.

references 335

Rainer, F. and Varela, S. (1992). ‘Compounding in Spanish’, Rivista di Linguistica

4: 117–42.

Ralli, A. (1992). ‘Compounding in Modern Greek’, Rivista di Linguistica 4:

143–74.

Rappaport, M., Levin, B., and Laughren, M. (1993). ‘Levels of Lexical Repre-

sentation’, in J. Pustejovsky (ed.), Semantics and the Lexicon, 37–54. Dordrecht:

Kluwer.

Rice, K. (1989). A Grammar of Slave. Berlin and New York: Mouton de Gruyter.

—— (2000). Morpheme Order and Semantic Scope: Word Formation in the

Athapaskan Verb. Cambridge: Cambridge University Press.

Riehemann, S. Z. (1998). ‘Type-Based Derivational Morphology’, Journal of

Comparative Germanic Linguistics 2: 49–77.

Riemsdijk, H. van (ed.) (1998). Clitics in the Languages of Europe. Berlin and New

York: Mouton de Gruyter.

Robins, R. H. (1959). ‘In Defense of WP’, Transactions of the Royal Philological

Society, 116–44.

Ronneberger-Sibold, E. (2000). ‘On Useful Darkness: Loss and Destruction of

Transparency by Linguistic Change, Borrowing andWord Creation’, YoM 1999,

97–120.

Rosen, S. T. (1989). ‘Two Types of Noun Incorporation: A Lexical Analysis’,Lg 65:

294–317.

Rubach, J., and Booij, G. (2001). ‘Allomorphy in Optimality Theory: Polish

Iotation’, Lg 77: 26–60.

Rumelhart, D., and McClelland, J. (1986). ‘On Learning the Past Tenses

of English Verbs: Implicit Rules or Parallel Distributed Processing?’, in

J. McClelland, D. Rumelhart, and the PDP Research Group (eds.), Parallel

Distributed Processing. Explorations in the Microstructure of Cognition, 216–71.

Cambridge: Cambridge University Press.

Ryder, M. E. (2001). ‘Complex -er Nominals: Where Grammaticalization and

Lexicalization Meet’, in E. Contini-Morava and Y. Tobin (eds.), Between

Grammar and Lexicon, 291–331. Amsterdam and Philadelphia: Benjamins.

Sadler, L., and Spencer, A. (2001). ‘Syntax as an Exponent of Morphological

Features’, YoM 2000, 71–96.

Salmon, V. (2000). ‘The Term ‘‘Morphology’’ ’, BLM i. 15–21.

Saussure, F. de (1916). Cours de linguistique generale. Paris: Payot.

Scalise, S. (1984). Generative Morphology. Dordrecht: Foris.

—— (1988). ‘The Notion of ‘‘Head’’ in Morphology’, YoM 1988, 229–46.

—— (1992a). ‘Compounding in Italian’, Rivista di Linguistica 4: 175–99.

—— (ed.) (1992b). ‘TheMorphology of Compounding’,Rivista di Linguistica 4 (1).

Schreuder, R., and Baayen, R. H. (1997). ‘How Complex Simplex Words Can

Be’, Journal of Memory and Language 37: 118–39.

336 references

Selkirk, E. O. (1982). The Syntax of Words. Cambridge, Mass.: MIT Press.

Shopen, T. (ed.) (1985). Language Typology and Syntactic Description, iii. Gram-

matical Categories and the Lexicon. Cambridge: Cambridge University Press.

SiptÆr, P., and T�rkenczy, M. (2000). The Phonology of Hungarian. Oxford:

Oxford University Press.

Song, J. J. (1996). Causativity and Causation: A Universal-Typological Perspective.

London and New York: Longman.

—— (2001). Linguistic Typology: Morphology and Syntax. Harlow: Longman.

Spencer, A. (1988). ‘Bracketing Paradoxes and the English Lexicon’, Lg 64:

663–82.

—— (1991). Morphological Theory. Oxford: Blackwell.

—— and Zaretskaya, M. (1998). ‘Stative Middles in Russian’, Essex Reports in

Linguistics, 22. Colchester: Dept of Language and Linguistics, University of

Essex.

—— and Zwicky, A. (eds.) (1998).AHandbook ofMorphology. Oxford: Blackwell.

Stroomer, H. (1987). ‘A Comparative Study of Three Southern Oromo Dialects in

Kenya’, Ph.D. diss. Univ. of Leiden.

Struijke, C. (2002). Existential Faithfulness: A Study of Reduplicative TETU,

Feature Movement, and Dissimilation. New York and London: Routledge.

Stump, G. (1993). ‘On Rules of Referral’, Lg 69: 447–79.

—— (2001). Inflectional Morphology: A Theory of Paradigm Structure. Cambridge:


Sulkala, H., and Karjalainen, M. (1992). Finnish. London and New York:

Routledge.

Szymanek, B. (1989). Introduction to Morphological Analysis. Warsaw: Panstwowe

Wydawnictwo Naukowe.

Talmy, L. (1985). ‘Lexicalization Patterns’, in Shopen 1985: 57–149.

Taylor, J. R. (2002). Cognitive Grammar. Oxford: Oxford University Press.

Tekoriene, D. (1990). Lithuanian: Basic Grammar and Conversation. Kaunas:

Spindulys.

ThrÆinsson, H. (1994). ‘Icelandic’, in Konig and van der Auwera 1994: 142–89.

Tiersma, P. (1982). ‘Local and General Markedness’, Lg 58: 832–49.

Uhlenbeck, E. M. (1978). Studies in Javanese Morphology. The Hague: Martinus

Nijhoff.

Unterbeck, B., Rissanen, M., Nevalainen, T., and Saari, M. (eds.) (2000).

Gender in Grammar and Cognition, 2 vols. Berlin and New York: Mouton De

Gruyter.

Verheij, A. C. J. (2000). Bits, Bytes, and Binyanim: A Quantitative Study of

Verbal Lexeme Formations in the Hebrew Bible. Leuven: Uitgeverij Peeters and

Departement Oosterse Studies.

references 337

Vigario, M. (2003). The Prosodic Word in European Portuguese. Berlin: Mouton de

Gruyter.

Vogel, I. (1994). ‘Verbs in Italian Morphology’, YoM 1993, 219–54.

Voorhoeve, J. (1979). ‘Multifunctionaliteit als derivationeel probleem’, in

T. Hoekstra and H. van der Hulst (eds.), Morfologie in Nederland, special issue

of GLOT (Leiden: Univ. of Leiden), 41–9.

Vries, L. de (1989). ‘Studies in Wambon and Kombai’, Ph.D. diss. University of

Amsterdam.

Weggelaar, C. (1986). ‘Noun Incorporation in Dutch’, International Journal of

American Linguistics, 52: 301–5.

Welmers, W. E. (1973). African Language Structures. Berkeley, Calif.: University

of California Press.

Wierzbicka, A. (1991). Cross-Cultural Pragmatics: The Semantics of Human

Interaction. Berlin: Mouton de Gruyter.

Wiese, R. (1996). The Phonology of German. Oxford: Clarendon Press.

Williams, E. (1981). ‘On the Notions ‘‘Lexically Related’’ and ‘‘Head of a Word’’’,

Linguistic Inquiry 12: 245–74.

Wiltshire, C., and Marantz, A. (2000). ‘Reduplication’, BLM i. 557–67.

Wurzel, W. U. (1980). ‘Ways of Morphologizing Phonological Rules’, in J. Fisiak

(ed.), Historical Morphology, 443–62. The Hague: Mouton.

Zwicky, A., and Pullum, G. (1983). ‘Cliticization vs Inflection: English n’t’, Lg 59:

502–13.

338 references

Language index

Afrikaans (South Africa) 59, 136, 202,210–11, 226, 259, 269–70, 275

Agta (Philippines) 47Alabama (Amerindian) 29Arabic 46

Bayso (Ethiopia) 126Bemba (Zambia) 120Biak (New Guinea) 174Bulgarian 46

Cree (Amerindian) 46

Diyari (Australia) 180–1, 200Dutch passimDyirbal (Australia) 193–4, 200

English passim

Finnish 101, 123, 126, 132–3, 147–8Flemish (Belgium) 202French 6, 19–22, 39, 45, 52, 83, 107–8,

111, 123, 144, 170, 181, 195, 261,266, 271, 273, 285

Frisian (The Netherlands) 92, 262–3,265

Gaelic 123Georgian 292German 7, 20–2, 29, 32, 37, 62, 64,

66, 68, 75, 80–2, 102–6, 109,112–13, 130, 137, 140, 169, 181,193, 224, 248, 257, 261, 264–4,267

Greek, Classical 19, 30, 33, 36, 179Greek, Modern 75, 88, 94, 286Greenlandic Eskimo 42, 194, 197

Hebrew, Biblical 61Hebrew, Modern 37–8, 107, 245Hungarian 75, 84, 106, 110, 132, 162,

183, 222

Icelandic 128–9, 167Ilokano (Philippines) 179

Italian 6, 28–9, 3–2, 34, 47, 55–6, 60, 73,77–9, 94, 113, 140–4, 166, 183, 203,253, 276

Ixil (Mexico) 196

Japanese 224Javanese (Java, Indonesia) 35

Kambera (Sumba, Indonesia) 110, 205Khmu (Laos) 29Kolyma Yukaghir (Siberia) 144

Latin 19, 30–1, 33, 39, 52, 55, 75, 80,100, 103–8, 111–12, 114, 142–3, 148,179, 261–2, 276, 289

Latvian 137Limburgian (The Netherlands) 181Lithuanian 134, 147

Malagasy (Madagascar) 73, 226Maori (New Zealand) 39, 78, 126

Nahuatl (Mexico) 197Ngiti (Congo) 12, 36, 128

Oromo (Ethiopia, Kenya) 25

Polish 15, 27–9, 32, 42, 99, 116–17,226

Ponapean (Micronesia) 92Portuguese 14, 290Punjabi (India, Pakistan) 81, 94–5

Quechua (South America) 60, 72, 139,165, 204–5

Russian 59, 110, 129

Sanskrit (India) 81, 196Sao-Tomense (Sao-Tome) 203Spanish 6, 21, 79, 90, 113, 136–7, 181,

285, 289Sranan 39Swahili (East-Africa) 130–1Swedish 171, 177, 277

Tagalog (Philippines) 47Turkana (Kenya) 44Turkish 42, 148–9, 200, 204–5, 226, 273Tuscarora (Amerindian) 198

Wambon (New Guinea) 15–16

Wolof (Africa) 197

Yidiny (Australia) 163Yucatec (Mexico) 94Yup’ik (Amerindian) 42, 208

340 language index

Index of terms

abessive 132ablative 132ablaut 36absolutive 192Absolute-Ergative system 192accusative 103acronym 20action nominalization 214action noun 214activation level 234activation spreading 235adessive 132adjectivalization 52adjunct 191adverbialization 52affix 9

competing 172phrasal 167postlexical 167

affix substitution 40affix telescoping 273affixoid 86Agent 91agrammatism 242agreement 82agglutinative language 42Aktionsart 135alignment 159allative 132alliterative concord 131allomorphy 31

paradigmatically governed 170alphabetism 20anaphora 187analogy 13analytic form 113Animacy Hierarchy 127anti-passive 196apophony 36applicative 197argument 91

core 190aspect 135

atelic 135durative 135grammaticall 135imperfective 135

lexical 135perfective 135predicational 135progressive 136telic 135

attenuative 15autosegmental morphology 179

back formation 40base-driven restriction 67base frequency 234base word 35bidirectional leveling 265binding 216binyan 61bleaching 266blend 20blocking 17borrowing 19bracketing paradox 212breaking 265

case 15direct 105inherent 105lexical 105local 132oblique 105semantic 105structural 105

category-changing 35category-determining 52category-neutral 54causative 199changeexternal 258internal 258

circumfix 29citation form 29class-changing 35clipping 20clitic 166coda 160combining form 30comitative 132comparative 101composition 75

compositional function 56Compositionality Principie 207compoundappositive 81bahuvrihi 80copulative 80dvandva 81endocentric 79exocentric 79neo-classical 30

compounding 5Compound Stress Rule 175concatenation 9conceptualization rule 311concord 107conflation 56conjugation 102connectionism 245constructional idiom 83construct state 107converb 102conversion 5cooperation 210co-phonology 177coreferentiality 15corpus 69correspondence rule 13cranberry morpheme 30creole language 259cumulative frequency 234cumulative root frequency 234cyclicity 175

dative 103declension 102deflection 260degrammaticalization 268degree 101deixis 134demotion 194dental preterite 266dependent marking 104dephonologization 263deponent verb 114derivation 5direction of 7

desinence 28desyntactization 263deverbalization 214diacritic feature 118dictionary 18diminutive 14Distributed Morphology 118domain shift 221drift 260

dual 44dual system 243

echo word formation 22elative 132ellipsis 20elliptical construction 192enclisis 166encyclopedic knowledge 209equative 101ergative 192evaluative morphology 14evidential 137exemplar-based model 248exclusive 139exponence 116cumulative 116extended 116

exponent 116

family size 235Faithfulness 158FinDevoicing 158finite 101frequency effect 234fusion 42fusional language 42fusion rule 119

gapping 164gender 108common 129feminine 129masculine 129neuter 129non-neuter 130

genitive 106genitivus objectivus 133genitivus subjectivus 133gerund 101gerundive nominal 214Goal 91Gesamtbedeutung 133government 104grammatical function 190grammaticalization 86

hapax 69head 53head marking 106head operation 77hierarchy 44honorific 224hypocoristic 21

342 index of terms

iconicity 59illative 132imperative 137imperfect learning 58inclusive 139indicative 137inessive 132infinite 101infinitive 101infix 29inflection 4

contextual 104inherent 104mixed 109strong 109weak 109

inflectional class 102inflectional ending 28inflectional homonymy 129inflectional paradigm 15inheritance 215inheritance tree 62input constraint 64interfix 88interlinear morphemic translation 41internal modification 36interpretation 209invisible hand theory 260irrealis 137isolating language 41Item-and-Arrangement

morphology 116Item-and-Process morphology 117

labelling function 14language contact 258lexeme 3lexeme formation 5Lexical Conceptual Structure 190lexical convention 17lexical decision 234lexical innovation 257Lexical Integrity 22lexicalization 17lexical norm 17lexical phonology 167lexicon 16

mental 18pan-European 19

linking element 88linking rule 191listeme 23

Main Stress Rule 175markedness 43

local 44meaning 209metaphony 37metaphorical use 221metonymy 221middle verb 39mixed category 214mood 137morph 31morpheme 8bound 9free 9lexical 9tonal 36

morpholexical rule 170morphological rule 10morphologically conditioned rule 169morphology 4concatenative 34evaluative 14lexeme-based 11morpheme-based 8non-concatenative 12

morphome 143morphophonology 33morphosyntactic category 99morphosyntactic feature 99motivated 207multifunctionality 39

neologism 232neural network 245nominalization 52nominative 103Nominative-Accusative system 192non-finite 101No Empty Onset Constraint 158Non-Redundancy Constraint 59No Phrase Constraint 188noun incorporation 92nucleus 160

obstruent 32onset 160Open Syllable Lengthening 169Optimality Theory 159output constraint 65overcharacterization 273overgeneralization 237

Panini’s Principle 118paradigm 8paradigmatic leveling 264paradigmatic relationship 8parser 5

index of terms 343

participle 101particle verb 186partitive 132passivization 194Patient 91paucal 126percolation 54perfect 136periphrasis 113person 138phrasal verb 22pidgin language 259pluralis majestatis 140polymorphemic 9polysemy 220polysynthetic language 41portmanteau morph 42position class morphology 119positive 101praesens historicum 134Predicate Argument Structure 190prefix 29preverb 202priming 240proclisis 166Pro Drop parameter 140productive 18productivity 67prosodic morphology 180prosodic structure 158prosody 65purist 257

qualifying adjective 211

radial category 223readjustment rule 119realis 137realizational morphology 117realization rule 117reanalysis 258recategorization 14Recipient 104redundancy rule 18reduplicant 178reduplication 35full 35partial 35

relational adjective 57relative frequency 250response latency 234result noun 215rhyme 160Right-hand Head Rule 54root 28

root-and-pattern-morphology 37rule of referral 129rule-governed creativity 6

Saussurean paradox 257schema 247scope 208separable complex verb 84serial verb 203shifter 138single system 245singulative 44sound symbolism 22spell out rule 117split ergativity 193split morphology 120stem 28stem allomorphy 88stratal constraint 65stratum 65strong lexicalism 120subcategorization 120subjunctive 137suffix 29cohering 155non-cohering 156

superlative 101suppletion 33suprafix 36suprasegmental 36switch reference 16syllable 160bimoraic 179

syllable structure 160syncretism 28syntactic valency 60syntagmatic relationship 8synthesis 41synthetic compounding 90synthetic form 113systematization 273

template 11templatic morphology 119tense 134future 134past 134present 134

thematic role 191thematic vowel 29TMA system 133token frequency 69transfix 37transcategorial construction 214transparency 34

344 index of terms

transposition 38explicit 38implicit 38

trial 126tripartite parallel structure 154truncation 21type coercion 67type frequency 69

umlaut 37underlying form 32unidirectionality 268univerbation 20universal

implicational 43markedness 43morphological 43

vagueness 210verbalization 52voice 196

active 196middle 196passive 196

vowel gradation 36vowel harmony 162vowel mutation 37

wordcomplex 7established 12formally complex 117grammatical 27orthographic 286phonological 154prosodic 158simplex 7

Word-and-Paradigmmorphology 117

word creation 20word family 4word form 3word-formation 5neo-classical 66paradigmatic 13parasynthetic 56

word grammar 13word manufacturing 20word recognition 235world knowledge 209

zero-ending 28zero-morpheme 39

index of terms 345

Documents

oxford textbooks in linguistics - 213.230.96.51:8090